Professional Documents
Culture Documents
Jayson L. Lusk is Professor and Willard Sparks Endowed Chair in the Depart-
ment of Agricultural Economics, Oklahoma State University.
Jason F. Shogren is Stroock Distinguished Professor of Natural Resource
Conservation and Management, and Professor of Economics and Finance,
University of Wyoming.
Quantitative Methods for Applied Economics and Business Research
Series Editor
PROFESSOR PHILIP HANS FRANSES
Erasmus University, Rotterdam
Researchers and practitioners in applied economics and business now have access to
a much richer and more varied choice of data than earlier generations. Quantitative
Methods for Applied Economics and Business Research is a new series aimed at meeting
the needs of graduate students, researchers and practitioners who have a basic grounding
in statistical analysis and who wish to take advantage of more sophisticated methodology
in their work.
Forthcoming titles
Stewart Jones and David Hensher (eds.), Credit Risk Modelling: A Primer
Experimental Auctions
Methods and Applications in Economic and
Marketing Research
www.cambridge.org
Information on this title: www.cambridge.org/9780521671248
C Jayson L. Lusk and Jason F. Shogren 2007
A catalogue record for this publication is available from the British Library
Lusk, Jayson.
Experimental auctions : methods and applications in economic and marketing
research / Jayson L. Lusk and Jason F. Shogren.
p. cm. – (Quantitative methods for applied economics and business research)
Includes bibliographical references and index.
ISBN–13: 978–0–521–85516–7 (hbk.)
ISBN–10: 0–521–85516–0 (hbk.)
ISBN–13: 978–0–521–67124–8 (pbk.)
ISBN–10: 0–521–67124–8 (pbk.)
1. Auctions. 2. Consumers’ preferences–Mathematical models. I. Shogren, Jason
F. II. Title. III. Series.
HF5476.L87 2007
330.072 4–dc22 2007002291
v
Contents
1 Introduction 1
1.1 Introduction 1
1.2 Why experimental auctions? 3
1.3 What is an experimental auction? 5
1.4 Purpose of this book and boundaries of coverage 17
3 Value theory 34
3.1 Introduction 34
3.2 Valuation under certainty 34
3.3 Valuation under uncertainty 37
3.4 Valuation in a dynamic environment with uncertainty, limited information,
and irreversibility 43
3.5 Summary 44
vii
viii Contents
6 Data analysis 95
6.1 Introduction 95
6.2 Censored regressions with auction bids 95
6.3 Quantile regression with auction bids 100
6.4 Panel data regression with auction bids 103
6.5 Other types of data analysis with auction bids 106
6.6 Conclusions 112
References 279
Index 297
Figures
ix
Tables
x
List of tables xi
1.1 Introduction
Our choices reflect our values. People reveal their relative values when they
choose to spend an extra hour at work rather than at the opera; purchase more
groceries rather than extra MP3s or drop extra change into a jar promoting
a charity at the check-out line rather than buying a candy bar. Economists
characterize the economic value of these choices by determining the rate at
which a person is willing to trade one good or resource for another. This rate is
captured in a person’s maximum willingness to pay to purchase a good or in their
minimum willingness to accept to sell a good. Usually, these economic values
are revealed within the context of an active exchange institution like a market
or auction with numerous buyers and sellers. In such exchange institutions,
buyers buy when their willingness to pay exceeds price and sellers sell when
their willingness to accept falls below price.
But, how do people value new goods and services not currently bought and
sold in the marketplace? These non-market goods and services include new
private goods like cigarettes that have been genetically modified to possess less
nicotine and diet cherry vanilla Coke with lime as well as public goods like
cleaner air in Santiago, Chile or biodiversity in Madagascar. No exchange insti-
tution exists for buyers and sellers to make bids and offers, which would reveal
people’s relative values for these non-market goods. But there are policymakers
and business managers who want information on the potential demand for these
goods; they want to know if the perceived benefits from the products outweigh
the costs to provide them.
Likewise, economists, psychologists, and marketers are also interested in
eliciting people’s values for both market and non-market goods. Economists
elicit values to conduct applied cost-benefit analysis related to public good
provision, and to estimate the welfare effects of technological innovation and
public policy (see e.g., Boardman et al., 2005). Psychologists and behavioral
economists want to learn about people’s values to understand the degree to
which decisions are consistent with preferences and beliefs and to offer refine-
ments to economic theory. This work focuses on how people’s values can be
1
2 Experimental Auctions
influenced by the context of the decision and how people use rules of thumb
to guide how they value goods (see Kahneman and Tversky, 2000). Marketing
experts are interested in eliciting values to better understand consumer prefer-
ences, forecast new product success, and measure effectiveness of promotional
activities, which in turn can help reduce the high failure rate of new products
and the significant costs of advertising (see Wertenbroch and Skiera, 2002).
Over the last four decades, researchers have developed many value elicita-
tion methods to tease out how people value various goods and services. These
methods can be broadly categorized as revealed or stated preference methods
(see Hanley et al., 2006 for an overview). Revealed preference methods use
existing market data to derive implicit values for a good, for example hedonic
pricing, travel costs. Revealed preferences work when the good already exists,
albeit indirectly, in the market. For example, while a natural wonder such as
the Grand Canyon cannot be directly bought and sold, we can observe how far
people drive and what they give up in terms of opportunity cost of time to visit
the Canyon. By detecting systematic patterns from these observations, one can
indirectly determine people’s value for the park. Another example: the number
of bathrooms in a house is not traded alone in the market; but by calculating the
difference in the sales price of a two-bathroom home and the sales price of an
otherwise identical one-bathroom home, we can indirectly determine people’s
values for an extra bathroom. The upside of revealed preference methods is
that real choices are examined. The downside of revealed preference methods
is that valuation is indirect and must be inferred from empirical patterns.
In contrast, stated preference methods use public opinion surveys or com-
parative choice trials that ask a person, directly or indirectly, to state his or her
value for the new good or service. The upside of stated preference methods
is that the researcher can create a hypothetical market where a person can, in
theory, buy or sell any good or service. The stated preference method is flexible
enough to construct alternative potential scenarios such that demand for the
good can be understood given changes in market and non-market conditions.
A well known downside of stated preference methods, regardless of how well
the survey is designed and executed, is that people know they are valuing a
hypothetical change in the good or service. The absence of market discipline,
which takes the form of budget constraints and availability of substitutes in
the real world, creates an environment conducive to questionable responses.
Values elicited from hypothetical surveys have exhibited many inconsistencies
such as a lack of responsiveness to the scale and scope of proposed benefits and
a tendency for people to promise to pay significantly more than they actually
do when asked to shell out the money, (see Diamond and Hausmann, 1994 and
Hanemann, 1994 for a discussion of the pros and cons of contingent valuation).
Traditional approaches used to elicit valuations suffer from several shortcom-
ings. Revealed preference methods are indirect and require several simplifying
Introduction 3
test theory because elicited values can be directly compared with the induced
value benchmark. This high level of experimental control, however, comes
at a cost. By definition, induced value experiments are abstract, focusing on
the allocative efficiency of the auction institution itself; these auctions do not
provide information on people’s values for real-world goods and services.
In response, researchers started applying what they learned in induced value
experiments to elicit people’s homegrown values: those values that people bring
into an experiment for real-world goods. Initial applications used experimental
auction-type mechanisms to elicit values for items such as public TV, sucrose
octa-acetate (a bitter liquid people bid to avoid tasting), and coffee mugs to
study the difference between willingness-to-accept and willingness-to-pay
measures of value and to determine people’s values for public goods (see Bohm,
1972; Coursey, Hovis, and Schulze, 1987; Kahneman et al., 1990). The work of
Hoffman et al. (1993) and Menkhaus et al. (1992) on the demand for vacuum-
packed meats was perhaps the first to use experimental auctions for marketing
purposes.
Today, experimental auctions are used around the world by applied
economists, psychologists, and marketers interested in valuing new products
and technologies and in investigating theoretical models of individual decision
making, auctions, and valuation. The reader can explore Table 1.1 to get a better
idea for how experimental auctions have been used over the last three decades.
Table 1.1 chronologically lists over 100 experimental auction studies. The list
helps illustrate the varied uses and expanding growth of experimental auctions
used to elicit valuations. These auctions have been used for a wide variety of
products. Applications range from valuing food safety (i.e., specific pathogens,
biotechnology, pesticides, traceability, and growth hormones), food attributes
(e.g., meat tenderness, meat color, fat content, and packaging), a variety of
foods (e.g., kiwis, apples, chocolates, potatoes, corn chips, cookies, milk, and
sandwiches), and a variety of non-food, high-value goods ranging from sports
cards to firm business records to used cars to gasoline to Christmas gifts.
Table 1.1 also shows that experimental auctions have been conducted for
a number of non-mutually exclusive reasons: to test theory, including inves-
tigations into the willingness-to-pay/willingness-to-accept divergence, studies
of preference reversals, tests of the commitment cost theory, and so on, to
study methods for valuing public and private goods, including investigations of
hypothetical bias, scope effects, the willingness-to-pay/willingness-to-accept
divergence, studies comparing mechanisms, studies of procedural issues, and
so on, and to elicit homegrown preferences, including preferences for risk and
time, and the demand for new goods and services.
When experimental auctions are used to elicit homegrown values, the
researcher aims to balance control and context. Control means the researcher
has control over the environment such that no unmeasured external force
drives choices. That is, confounding of cause and effect is eliminated. What
7
Table 1.1. Examples of experimental auctions in action
1 Becker, DeGroot, and Marschak Lotteries Estimate risk preferences USA Behavioural Science
1964
2 Bohm Public television show Study methods for valuing Sweden European Economic Review
1972 public goods
3 Grether and Plott Lotteries Study preference reversal USA American Economic Review
1979 phenomenon
4 Bennett Movie Study contributions to a public Australia Economic Analysis and Policy
1983 good
5 Bohm Bus route Study methods for valuing Sweden Public Finance and the Quest for
1984 public goods Efficiency
6 Cummings, et al. Sucrose octa-acetate (a bitter Study methods for valuing USA Valuing Environmental Goods:
1986 tasting liquid), public public goods An Assessment of the CVM
goods
7 Brookshire and Coursey Tree density in a park Study methods for valuing USA American Economic Review
1987 public goods
8 Brookshire et al. Strawberries Study external validity of USA Economic Inquiry
1987 experimental auctions
9 Coursey et al. Sucrose octa-acetate (a bitter Study WTP/WTA divergence USA Quarterly Journal of Economics
1987 tasting liquid)
10 1988 Loewenstein Delayed cash payment Estimate time preferences USA Management Science
11 Harless Lotteries Study WTP/WTA divergence USA Journal of Economic Behavior
1989 and Organization
12 Kahneman, Knetsch, and Thaler Coffee mugs Study WTP/WTA divergence USA Journal of Political Economy
1990
13 Shogren Protection and insurance Study risk reduction USA Journal of Risk and Uncertainty
1990 against monetary loss mechanisms
14 Crocker and Shogren Lotteries Study preference learning USA Environmental Policy and the
1991 Economy
(cont.)
8
Table 1.1. (cont.)
15 Boyce et al. Life of small pine tree Study WTP/WTA divergence USA American Economic Review
1992
16 1992 Kachelmeier and Shehata Lotteries Study WTP/WTA divergence; China American Economic Review
estimate risk preferences
17 1992 Menkhaus et al. Beef steaks Estimate determinants of value USA Journal of Agricultural and
for vacuum packaging Resource Economics
18 Buhr et al. Pork sandwich Value growth hormones and USA Journal of Agricultural and
1993 marbling Resource Economics
19 Hoffman et al. Beef steaks Study procedural issues; value USA Marketing Science
1993 vacuum packaging
20 McClelland et al. Insurance to avoid loss Study risk preferences; study USA Journal of Risk and Uncertainty
1993 hypothetical bias
21 Bohm Used cars Study preference reversal Sweden Empirical Economics
1994 phenomenon
22 Fox et al. Milk Value growth hormones USA Journal of Dairy Science
1994
23 Shogren and Crocker Protection and insurance Study preferences for timing of USA Economics Letters
1994 against monetary loss risk reduction
24 Shogren et al. Candy bars, pork sandwiches Study WTP/WTA divergence USA American Economic Review
1994
25 Fox et al. Pork sandwiches Value growth hormones USA Journal of Animal Science
1995
26 Hayes et al. Pork sandwiches Value food safety USA American Journal of
1995 Agricultural Economics
27 Di Mauro and Maffioletti Protection and insurance Study preferences for Italy Journal of Risk and Uncertainty
1996 against monetary loss ambiguity
28 Melton et al. Pork chops Value meat color, marbling, USA American Journal of
1996 size, and tenderness Agricultural Economics
29 Bateman et al. Gourmet chocolates, soft Study WTP/WTA divergence UK Quarterly Journal of Economics
1997 drink
9
30 Bohm et al. 30 liters of gasoline Study procedural issues; Sweden Economic Journal
1997 compare mechanisms
31 Frykblom Atlas Study hypothetical bias Sweden Journal of Environmental
1997 Economics and Management
32 Kirby Delayed cash payment Estimate time preferences USA Journal of Experimental
1997 Psychology: General
33 List and Shogren Sports cards Study hypothetical bias USA Journal of Economic Behavior
1998 and Organization
34 List and Shogren Various Christmas gifts Estimate deadweight loss of USA American Economic Review
1998 Christmas
35 List et al. Sports cards Study hypothetical bias USA Economics Letters
1998
36 Roosen et al. Apples Value pesticide use USA Journal of Agricultural and
1998 Resource Economics
37 1998 Rutström Gourmet chocolates Compare mechanisms USA International Journal of Game
Theory
38 1998 Fox et al. Pork sandwiches Study hypothetical bias USA American Journal of
Agricultural Economics
39 List and Shogren Candy bars, pork sandwiches Study effect of price feedback USA American Journal of
1999 on bids Agricultural Economics
40 Lucking-Reiley Trading cards Compare mechanisms in USA American Economic Review
1999 on-line auctions
41 Frykblom and Shogren Atlas Study methods for valuing Sweden Environmental and Resource
2000 public goods Economics
42 Horowitz and McConnell Binoculars, coffee mugs, Study hypothetical bias; study USA Journal of Economic Behavior
2000 flashlights performance of mechanism and Organization
43 List and Lucking-Reiley Sports cards Compare mechanisms USA American Economic Review
2000
44 Shogren, List, and Hayes Candy bars, mangos, pork Test for preference learning vs. USA American Journal of
2000 sandwiches experimental novelty Agricultural Economics
(cont.)
10
Table 1.1. (cont.)
45 Balistreri et al. Lotteries Study hypothetical bias USA Environmental and Resource
2001 Economics
46 2001 Knetch et al. Coffee mugs Study WTP/WTA divergence Canada, Experimental Economics
Singa-
pore
47 List Sports cards Study methods for valuing USA American Economic Review
2001 public goods
48 Lusk et al. Beef steaks Value tenderness USA American Journal of
2001 Agricultural Economics
49 Lusk et al. Corn chips Value genetically modified USA Journal of Agricultural and
2001 food Resource Economics
50 Shogren et al. Candy bars, coffee mugs Study WTP/WTA divergence USA Resource and Energy Economics
2001
51 Dickinson and Bailey Beef sandwiches, pork Value traceability, food safety, USA Journal of Agricultural and
2002 sandwiches production methods Resource Economics
52 Fox et al. Pork sandwiches Study effect of information USA Journal of Risk and Uncertainty
2002 about irradiation
53 Huck and Weizäcker Contracts tied to other Study people’s ability to Germany Journal of Economic Behavior
2002 people’s choices predict others’ preferences and Organization
54 Lange et al. Champagne Study performance of France Food Quality and Preference
2002 mechanism
55 List Sports cards Study preference reversal USA American Economic Review
2002 phenomenon
56 Masters and Sanogo Infant foods Estimate welfare effects of Mali American Journal of
2002 quality certification Agricultural Economics
57 Noussair, Robin, and Ruffieux Corn flakes Study effects of labels on France Economics Letters
2002 genetically modified food
58 Soler and Sanchez Vegetables Value organic and eco labels Spain British Food Journal
2002
59 Umberger et al. Beef steaks Value corn fed vs. grass fed USA Agribusiness
2002 beef
11
60 Wertenbroch and Skiera Cake, pen, soft drink Study performance of Germany, Journal of Marketing Research
2002 mechanism USA
61 Alfnes and Rickertsen Beef steaks Value growth hormones and Norway American Journal of
2003 country of origin Agricultural Economics
62 2003 Areily, Loewenstein, and Prelec Annoying sounds, keyboard, Test theory of coherent USA Quarterly Journal of Economics
wine arbitrariness
63 Cherry, Crocker, and Shogren Monetary and wildlife Study preference reversals USA Journal of Environmental
2003 lotteries Economics and Management
64 2003 Hong and Nishimura Lotteries Compare mechanisms; study USA Journal of Economic Behavior
mechanism performance and Organization
65 Huffman et al. Corn chips, potatoes, Value genetically modified USA Journal of Agricultural and
2003 vegetable oil food Resource Economics
66 List Sports cards Study WTP/WTA divergence USA Quarterly Journal of Economics
2003
67 Loureiro, Umberger, and Hine Cookies Study procedural issues USA Applied Economics Letters
2003
68 Lusk Coffee mug, lotteries Test commitment cost theory USA American Journal of
2003 Agricultural Economics
69 Stoneham, Chaudhri, and Land conservation contracts Value biodiversity and Australia Australian Journal of
2003 Strappazzon conservation; test-bed Agricultural and Resource
mechanism Economics
70 Umberger et al. Beef steaks Value country of origin USA Journal of Food Distribution
2003 Research
71 Blondel and Javaheri Apples, wine Value organic food France Acta Horticulturae
2004
72 Carpenter, Holmes, and Over 20 items including Compare mechanisms in USA IZA Discussion Paper
2004 Matthews DVD players, gift charity auctions
certificates, and toys
73 Cummings, Holt, and Laury Water permits Value irrigation rights USA Journal of Policy Analysis and
2004 Modeling
74 Feuz et al. Beef steaks Value tenderness and flavor USA Journal of Agricultural and
2004 Resource Economics
(cont.)
12
Table 1.1. (cont.)
75 2004 Hofler and List Sports cards Study hypothetical bias USA American Journal of
Agricultural Economics
76 Killinger et al. Beef steaks Value color, marbling, and USA Journal of Animal Science
2004 origin
77 List Sports cards Study discrimination USA Quarterly Journal of Economics
2004
78 Lunander and Nilsson Contracts for road painting Study mechanism; test-bed Sweden Journal of Regulatory
2004 mechanism Economics
79 Lusk et al. Beef steaks Value meat quality; compare USA American Journal of
2004 mechanisms; test for Agricultural Economics
endowment effects
80 Lusk et al. Cookies Study effect of information France, European Review of
2004 about biotechnology UK, Agricultural Economics
USA
81 Nalley et al. Sweet potatoes Value taste, origin, and health USA Mississippi State University
2004
82 Noussair et al. Biscuits Value genetically modified France Economic Journal
2004 food; investigate effect of
tolerance levels
83 Noussair et al. Candy bars, cookies, orange Study performance of France Food Quality and Preference
2004 juice mechanism
84 2004a Rousu et al. Corn chips, potatoes, Study effect of genetically USA Review of Agricultural
vegetable oil modified food tolerance Economics
limits
85 2004b Rousu et al. Corn chips, potatoes, Value conflicting information USA Land Economics
vegetable oil on genetically modified food
86 Rozan et al. Apples, bread, potatoes Value metal content; compare France European Review of
2004 mechanisms Agricultural Economics
87 Umberger and Feuz Beef steaks Study performance of USA Review of Agricultural
2004 mechanism Economics
13
88 Ackert et al. Mugs Study WTP/WTA divergence USA Federal Reserve Bank of Atlanta
2005
89 Berg et al. Lotteries Estimate risk preferences; USA Proceedings of the National
2005 compare mechanisms Academy of Sciences
90 Bernard Chocolates Study effect of price feedback USA Applied Economics Letters
2005 on bids; value organic food
91 Bernard and Schulze MP3 player Study how people forecast USA Economics Bulletin
2005 future values
92 Brown et al. Chicken sandwich Value food safety Canada Canadian Journal of
2005 Agricultural Economics
93 Corrigan Coffee mug Test commitment cost theory USA Environmental and Resource
2005 Economics
94 Dickinson and Bailey Beef sandwiches, pork Value traceability, food safety, Canada, Journal of Agricultural and
2005 sandwiches production methods Japan, Applied Economics
UK,
USA
95 Ding et al. Chinese food meals Study performance of USA Journal of Marketing Research
2005 mechanism
96 Hobbs et al. Beef sandwiches, pork Value traceability, food safety, Canada Canadian Journal of
2005 sandwiches production methods Agricultural Economics
97 Hudson, Coble, and Lusk Lotteries Estimate risk preferences USA Agricultural Economics
2005
98 Jaeger and Harker Kiwi fruit Value new kiwi variety; value New Journal of the Science of Food
2005 genetically modified food Zealand and Agriculture
99 Kassardjian et al. Apples Value genetically modified New British Food Journal
2005 food Zealand
100 Lusk et al. Cookies Estimate welfare effects of France, Economics Letters
2005 biotechnology policies UK,
USA
101 Platter et al. Beef steaks Value meat color, marbling, USA Journal of Animal Science
2005 size, and tenderness
102 Plott and Zeiler Lotteries and mugs Study WTP/WTA divergence USA American Economic Review
2005
(cont.)
14
Table 1.1. (cont.)
103 2005 Rousu et al. Cigarettes Value genetically modified USA Journal of Agricultural and
cigarettes with quality Applied Economics
improvement
104 Cherry and Shogren Lotteries Study preferences with USA Journal of Economic
2006 market-like arbitrate Psychology
105 Corrigan and Rousu Corn chips, salsa Test for endowment effects USA American Journal of
2006 Agricultural Economics
106 Corrigan and Rousu Candy bars, coffee mugs Study effect of price feedback USA American Journal of
2006 on bids Agricultural Economics
107 Eigenraam et al. Land conservation contracts Value biodiversity and Australia Department of Primary
2006 conservation; test-bed Industries
mechanism
108 Hobbs, Sanderson, and Haghiri Bison meat sandwich, beef Value bison meat; value health Canada Canadian Journal of
2006 sandwich information Agricultural Economics
109 Lusk et al. Cookies Value genetically modified France, Agricultural Economics
2006 food UK,
USA
110 Marcellino Business records Value farm records USA Purdue University
2006
111 Marette et al. Fish Value omega 3 fatty acid and France Iowa State University
2006 metal content
112 Norwood and Lusk Soft drinks Test theory of excessive choice USA Oklahoma State University
2006 effect
113 Shaw, Nayga, and Silva Cookie Value information on health USA Economics Bulletin
2006 risk
Introduction 15
weakly dominant strategy (i.e., a strategy that yields at least as good an outcome
as any other) is to bid his or her real value for the good. Chapter 2 discusses
the incentive properties of the second price auction in more detail, as well as
several other incentive-compatible mechanisms including, the BeckerDeGroot-
Marschak mechanism, the English auction, Vickrey nth price auctions, and the
random nth price auction. These auctions vary in terms of the level of market
interaction and feedback, number of winners and the market price, but all yield
the same result in theory.
2.1 Introduction
Auctions have been used for centuries as a price-discovery mechanism
(Lucking-Reiley, 2000). The theoretical study of auctions, however, is a rela-
tively recent phenomenon. Starting with the pioneering work of William Vick-
rey in 1961, economists have developed a rich literature devoted to auction
theory which is astounding in its results such as the revenue equivalence the-
orem and in its growing complexity. In addition to inventing and studying the
properties of alternative auction mechanisms, theorists have addressed issues
related to how bidding behavior is affected by numbers of bidders, informa-
tion, value uncertainty, risk preferences, violations of expected utility theory,
value interdependence, asymmetry, and multiple-unit demand. Good reference
books and papers include: Klemperer (1999, 2004), Krishna (2002), Milgrom
(2004), and McAfee and McMillan (1987). Such texts focus primarily on devel-
oping and espousing particular theoretical properties of auctions, with attention
devoted to designing auctions for generating maximum possible revenue or effi-
ciency (e.g., the ability of an auction to allocate units to the person or people
with the highest value(s) for the auctioned good(s)). Students of experimental
auctions should not by-pass this literature as it is important to understand such
topics as revenue equivalence and efficiency.
Our book purposefully restricts the theoretical treatment of auctions to the
question of incentive compatibility. An auction is said to be incentive compatible
when it induces each bidder to submit a bid that sincerely reflects his or her
value for the good. Auctions that elicit such truthful value statements focus the
reader on the key issue at hand: eliciting values for use in cost/benefit analysis,
marketing research, and testing the behavioral foundations of microeconomic
theory.
A variety of auctions are considered incentive compatible, meaning each
bidder has a weakly dominant strategy to submit a bid equal to their value.
These auctions can be identified in that they separate what people say from
what they pay. That is, an auction mechanism is incentive compatible if the
market price paid by a person is independent from what he or she bids. The
19
20 Experimental Auctions
1 In our discussion, we use the term incentive compatibility to refer to the theoretical properties
of a mechanism, while the term demand revealing is used to refer to the empirical properties of
a mechanism.
Incentive compatible auctions: theory and evidence 21
function increasing in income. If the bidder does not win the auction, his or
her monetary value from bidding is normalized to zero. At the time the bid
is submitted, the bidder does not know the second highest bid and thus does
not know the price that will be paid. In effect, the price is a random variable.
Suppose bidder i’s expectation about the price is characterized by the cumulative
distribution function G i ( p) with support [ pi , pi ] and the associated probability
density function gi ( p). The goal of the bidder is to submit a bid, bi , to maximize
expected utility, which is given by:
bi pi
E[Ui ] = Ui (vi − p)dG i ( p) + Ui (0)
pi bi
bi pi
= Ui (vi − p)gi ( p)d p + Ui (0). (2.1)
pi bi
The first integral is taken over all price levels less than his or her bid: cases in
which the bidder wins the auction. The second integral is taken over all price
levels greater than the bidder’s bid: cases in which he or she loses the auction.
Normalizing U (0) = 0, we find the optimal bid by taking the derivative of
expression (2.1) with respect to bi and setting the derivative equal to zero which
yields:
∂ E[Ui ]
= Ui (vi − bi )gi (bi ) = 0. (2.2)
∂bi
Equation (2.2) is solved when bi = vi . The bidder’s expected utility is maxi-
mized when he or she submits a bid equal to his or her value for the good. This is
a general finding for the second price auction. The optimal strategy of “bidding
one’s value” does not depend on the bidder’s risk preferences, the number of
rival bidders, initial wealth levels, or any of the other bidders’ bidding strate-
gies. Interested readers can consult Milgrom and Weber’s (1982) theorem 6 for
a rigorous proof of the incentive compatibility of the second price auction.
There is a more intuitive way to illustrate the incentive properties of the
second price auction. If a bidder submits a bid greater than his value, he runs
the risk that the second highest bid will exceed his value, which would cause
him to lose money; if he submits a bid less than his value, he runs the risk that
someone could outbid him, causing him to miss out on a profitable opportunity.
Over- and under-bidding one’s value runs the risk of either paying too much or
missing out on a good deal, which drives the bidder toward simply bidding his
true value. By separating what a person pays from what they say, the second price
auction induces sincere bidding in theory. We illustrate this intuition below.
Again, let vi be a bidder’s value for a good, bi be their bid, p be the market
price, and Ui be a strictly increasing utility function. By definition if bi > p,
22 Experimental Auctions
Bidding strategy
Under-bid
Realized price (bi < vi ) Over-bid (bi > vi ) Bid value (bi = vi )
p > vi > bi 0
vi > p > bi 0
vi > bi > p Ui (vi − p) > 0
p > bi > vi 0
bi > p > vi Ui (vi − p) < 0
bi > vi > p Ui (vi − p) > 0
p > vi = bi 0
vi = b > p Ui (vi − p) > 0
then a bidder wins the auction and pays p. In contrast, if bi ≤ p, the bidder does
not win the auction and pays nothing. Each bidder has three possible bidding
strategies: bid less than true value (bi < vi ), bid more than true value (bi > vi ),
or bid true value (bi = vi ). Table 2.1 shows the possible payoff outcomes that
could result from the three bidding strategies given various realizations of the
market price, p.
First, assume the bidder underbids (i.e., bi < vi ). Table 2.1 shows that three
possible payoffs can be realized: (i) If p > vi > bi , the bidder loses the auction
and earns 0; (ii) if vi > p > bi , the bidder again loses and earns 0; however, note
that positive utility could have been experienced had they bid higher; and (iii) if
vi > bi > p, the bidder wins the auction, pays p, and realizes Ui (vi − p) > 0.
Assume now the bidder overbids (i.e., bi > vi ). Again, three outcomes are
possible: (i) If p > bi > vi , the bidder loses and earns 0; (ii) if bi > p > vi ,
the bidder wins the auction, but because p > vi , realized utility is negative:
Ui (vi − p) < 0, the bidder would have been better off decreasing their bid; and
(iii) if bi > vi > p, the bidder wins, pays p, and realizes Ui (vi − p) > 0.
Finally, suppose the bidder pursues a truthful bidding strategy (i.e., bi = vi ).
Two outcomes are possible: (i) if p > vi = bi , the bidder loses and earns 0, or
(ii) if vi = bi > p, the bidder wins, pays p, and realizes Ui (vi − p) > 0.
The question is whether our bidder would have faired better if they had
bid their true value rather than under- or over-bidding. Table 2.2 shows the
outcomes by taking the difference between utility when the bidder bid their
true value and the utility when they under- or over-bid. In most cases, the
expected payoff is the same regardless of whether the bidder bids true value
or under- or over-bid. In two cases, however, bidding true value generates
a better outcome than under- or over-bidding. For under-bidding, the bidder
could have won the auction and experienced positive utility if only a higher bid
were offered. For over-bidding, the bidder could have prevented a loss yielding
Incentive compatible auctions: theory and evidence 23
Table 2.2. Payoff from bidding true value instead of under- or over-bidding
Realized price
> Under-bidding (bi < vi ) Over-bidding (bi > vi )
negative utility if only a lower bid were offered. The strategy of bidding one’s
true value never lowers utility and can increase utility relative to under- or
over-bidding. Stated differently, a bidder can never generate a higher level of
utility, but might generate less by offering bids that diverge from their valuation.
Therefore, the weakly dominant strategy in the second price auction is to bid
one’s true value.
The key behind the incentives of the second price auctions and similar incen-
tive compatible mechanisms is that the price paid by the bidder, p, is unaffected
by their bid, bi . To illustrate this point more forcefully, it is instructive to com-
pare the bidding strategy in the second price auction to another mechanism in
which the bid influences the price, such as the first price auction. In a first price
auction, individual i submits a bid, bi , to obtain a good against N rival bidders.
The bidder wins the auction if they submit the highest bid and this winning
bidder pays a price equal to their own bid. The optimal bidding strategy now
depends on what the bidder believes other people will bid and how many other
people participate in the auction. In a first price auction, a bidder weighs two
competing factors: increasing one’s bid increases the chance of winning, but
it also implies lower utility because one must pay a higher bid. All bidders
weigh these two competing factors when determining their best response to
their rivals’ best responses. As shown by Vickrey (1961) and others, if bidders
are symmetric and risk neutral, with values drawn from a uniform distribution
with support [v, v], the closed form solution for a Nash equilibrium bid function,
biNE , is:
N −1
biNE = v + (vi − v). (2.3)
N
If v = 0, the Nash equilibrium bid is (N − 1)/N times the bidder’s value, vi .
In a first price auction, people do not have an incentive to reveal their true
value. Rather they shave their bids downward based on the number of bidders
in the auction. In a two person auction, for instance, a bidder would submit a
bid equal to half their value, bi = 1/2vi . One might be tempted to draw the
24 Experimental Auctions
conclusion that a first price auction will provide a good approximation to true
values if N is large; however, one must recognize the result in expression (2.3)
also depends on the strong assumptions of a uniform distribution for values,
risk neutrality, and symmetry. A violation of any of these assumptions can alter
the Nash equilibrium bid function.
Of course the weakly dominant strategy of truthful value revelation in the
second price auction rests on assumptions as well. These assumptions are sig-
nificantly weaker than those required to arrive at a closed form solution for
the first price auction. That said, one still should recognize that situations arise
in which the second price auction and similar mechanisms are not incentive
compatible. This is the topic of the next section.
on to prove that there is no mechanism for eliciting true values for lotteries
that does not depend on the independence axiom. Popular variants of expected
utility theory (e.g., rank-dependent utility theory in Quiggin (1982)) relax the
independence assumption, implying that it is a difficult task to determine a
bidder’s value for a lottery once one foregoes the assumption of expected utility
theory. Karni and Safra (1987) show this result pertains beyond the BDM to
the second price auction, and by implication to all nth price auctions. Hong and
Nishimura (2003) showed that these theoretical results go beyond academic
curiosity. In an experimental study, they found that prices from English auctions
for lotteries significantly exceeded prices from second price auctions, which
according to the theoretical results of Neilson (1994), implies people exhibited
non-expected utility preferences when bidding in the second price auction.
Still, direct tests of expected utility theory relative to competing theories sug-
gest that expected utility may not be an unreasonable assumption. Hey and Orme
(1994) and Harless and Camerer (1994) provide tests between the major com-
peting theories. Both studies found that no single theory best described behavior
and that there was significant heterogeneity in preference functions, at least for
the college students used as subjects. Hey and Orme (1994) found expected
utility theory provided the best fit to the experimental data with rank-dependent
expected utility theory coming in second. Harless and Camerer (1994) inves-
tigated the trade off between parsimony and model fit and developed a menu
of theories that could be used in this regard including mixed fanning, prospect
theory, expected utility theory, and expected value. On a final note, these cri-
tiques relate to the valuation of random goods or those for which people are
uncertain of the value.2
Now consider the question of affiliated values. Attention has focused on
investigating bidding behavior when people have interdependent preferences.
If the value to one bidder depends in part on information available to other
bidders (i.e., values are affiliated), Milgrom and Weber (1982) and McAfee and
McMillan (1987) show that the second price auction is neither incentive com-
patible nor efficient. Klemperer (1999) provides a straightforward illustration
of the problem. Consider the general case in which a bidder’s value for a good
is given by vi = αti + β j=i t j , where t is a signal received by each bidder. If
β = 0, the model collapses to the case of private independent values. If α = β,
2 Horowitz (2005) suggests violations of expected utility theory could affect how people reveal
values for non-random goods (i.e., those goods for which individuals are certain about the value).
If people exhibit a certain type of non-expected utility preference function, as in Machina (1982),
the optimal bid in a BDM mechanism or second price auction need not equal a bidder’s value.
Machina’s preference function is characterized by people possessing a local utility function that
depends on the particular probability distribution – e.g., Ui (vi |G). While Horowitz’s results imply
that the incentive compatibility of the BDM and second price auction are not universal, his results
are not a condemnation of value elicitation for all non-expected utility models. For example, the
BDM and nth price auctions are incentive compatible if individuals have rank-dependent expected
utility or disappointment aversion (Gul 1991) preferences.
26 Experimental Auctions
the model captures the idea of pure common values, that is everyone values the
good identically. In reality, most auctions can be characterized by α > β, where
people combine their own signal of value with others’ values given imperfect
information about quality, resale value, or concerns related to social standing.
Because individual i does not know others’ signals, they do not know the precise
value of the good to themselves. The bidder must form some expectation of the
signal of all other bidders. Assuming bidders are risk neutral and signals are
independently drawn from a uniform distribution on [0, t], Klemperer (1999)
shows the optimal bid in a second price auction is:
∗ N
bi = α + β ti . (2.4)
2
A bidder will not bid his true value under affiliation; he could over- or under-bid
depending on the number of bidders and the magnitude of β.
To get around the problem of affiliation, Dasgupta and Maskin (2000) pro-
posed a generalization of the Vickrey auction in which each bidder submits a
bid function contingent on other bidders’ private values. In their efficient auc-
tion, the auctioneer takes these contingent bids and calculates each bidder’s
bid as a function of the bid functions, then determines the high bidder and the
price, which is also a function of all bid functions other than the high bidder.
Given affiliation, the key is that the auction once again separates what bid-
ders pay from what they say, such that efficiency is achieved: the bidder with
highest value wins the auction. Although some empirical support for Dasgupta
and Maskin’s efficient mechanism has been found (e.g., Margolis and Shogren,
2004), its practical implementation could prove difficult as it requires each
bidder to submit bid functions over all possible preferences of rivals.
Perry and Reny (2002) suggest an alternative two-step bidding mechanism
that requires each bidder to submit a bid for a good (or goods) in an initial
round and then requires each bidder to submit a bid against each and every
other bidder in a second round. Again, the mechanism separates what people
say from what they pay. As noted by Perry and Reny (p. 1207), “[b]idder i’s
second-round bids against bidder j are submitted after i infers the entire vector
of signals from the first-round bids. However, i’s bids against j are independent
of j’s first-round bids. Consequently, j is unable to affect the price he pays for
any unit.”
Finally, we note that in deriving the incentive compatibility property of the
second price auction, the utility function was defined only over income and
goods consumed in the experiment. A person may not strictly bid true value
if they have goals that extend beyond the immediate experimental context. For
example, people may derive utility from sending a “signal” to researchers to
influence future prices or offerings of a good or service that might later be
purchased in the marketplace. Also, a person might derive utility from “looking
Incentive compatible auctions: theory and evidence 27
good” to the researcher and this type of social desirability bias might manifest
itself in bidding behavior. Bidders might just like to win for winning’s sake –
to be the top dog. However, virtually all experimental economic and marketing
research methods are open to such criticism. The key point is that incentive
compatible mechanisms, such as the second price auction, impose an immediate
cost on people if they choose to send “signals” by bidding in a manner that
deviates from their true value. The opportunity cost of non-optimal bidding is
immediately obvious in the form of negative profits.
This section emphasizes that the class of auctions we consider in this book
are incentive compatible given a set of assumptions. The reader will have to
determine whether the assumptions underlying the theory of incentive compat-
ibility holds for the good and context in which their auction is conducted. We
considered three issues which might render an auction non-incentive compati-
ble. First, if one believes bidders are not expected utility maximizers, standard
auctions for goods with uncertain outcomes might not yield the desired bidding
behavior. Thus far, the literature has yet to offer many solutions for how to
truthfully elicit valuations in an auction context, when individuals have non-
expected utility preferences. One suggestion offered by Tversky et al. (1990)
is to have people bid on goods (or lotteries), but determine allocations based
only on the ranking (and not the magnitude) of bids. Second, if one believes
that bidders do not have independent private values, bidding behavior might
not be in accordance to theory. Here several mechanisms exist, such as those
suggested by Dasgupta and Maskin (2000) or Perry and Reny (2002), which can
restore the incentive properties of the auctions. Third, people might not bid true
value if they attempt to please the researcher with their bidding behavior or
if they believe their response to be consequential beyond the immediate auc-
tion context. One way to better understand the robustness of the auction theory
assumptions and concepts is to test them in an empirical setting. This is the
topic of the next section.
bidding behavior. The results from these studies suggested the second price
auction was not demand revealing, rather people tended to over-bid in second
price auctions. For example, Kagel et al. (1987) found that in 80% of auctions
investigated, the market price exceeded the second highest induced value in the
second price auction. This finding contrasted sharply with the English auction in
which 76% of bidding behavior was characterized by the truth-telling strategy.
Despite these findings, Kagel (1995) contended that it is surprising that devia-
tions from the truthful bidding strategy were not larger than what was observed,
given the lack of transparency of truthful bidding when such a strategy is not
revealed, and that (p. 513), “the dominant bidding strategy does have some
drawing power in explaining behavior . . . Thus, what we are seeing in the
second-price auction is relatively stable and modest (in terms of expected cost)
bidding above valuations, rather than a complete collapse of the theory.”
The influence of the findings from Kagel et al. (1987) and their on-going
work cannot be understated. Their findings have instilled a commonly held view
among economists that the second price auction is not demand revealing and
that the typical result from such auctions is over-bidding relative to valuations.
For example, in a recent paper Bernard (2005) indicates, “One of the more
commonly known and used results of auction theory is the equivalence of
the second-price and English auction mechanisms . . . Despite the theoretical
equivalence of the two auctions, experimental studies have shown consistently
that only the latter matches predictions. Instead, many studies . . . have found a
strong tendency for subjects in a second-price auction to submit bids great than
their values.”
Unfortunately, this view does not consider more recent papers that focused
on individual bidding behavior, in which the hypothesis of truthful bidding in
the second price auction cannot be rejected (e.g., Noussair, Robin, and Ruffieux,
2004; Shogren et al., 2001a; Parkhurst, Shogren, and Dickinson, 2004; Lusk
and Rousu, 2006). These papers typically estimate the parameters (α and β)
of the following bid function: bid = α + β (induced value) and test the joint
hypothesis that α = 0 and β = 1.
In many ways, this analysis is more convincing than earlier studies as it
relates to whether the market price is as expected and whether all bids are as
expected. Even if one looks at individual bidding behavior and does not address
the estimated bid functions, it becomes evident that over-bidding, at the indi-
vidual bidder level, is not necessarily the norm in second price auctions. For
example, Shogren et al. (2001a) found that of 280 second price auction bids,
about 56% were exactly equal to induced values, 33% were less than induced
values, and only 11% were over-bids; further 65% all bids were within $0.10
of induced values. Similarly, Lusk and Rousu (2006) found that about 38%
of subjects in their induced value study bid exactly true value after only two
rounds of learning (68% bid within two tokens of their induced value) while
30 Experimental Auctions
only 33% over-bid. Further, although Hong and Nishimura (2003) were, in
some circumstances, able to reject the equality of market prices and the second
highest bid, they observed under-bidding in the second price auction. Nous-
sair et al. (2004) also observed significant under-bidding in the second price
auction in initial bidding rounds. At this point it is unknown exactly why the
more recent results of Hong and Nishimura (2003), Shogren et al. (2001a),
Parkhurst et al. (2004), and others differ from the earlier results in Kagel
et al. (1987), Kagel and Levin (1993), and Harstad (2000), although differences
in instructions, experimental methods, and methods of data analysis certainly
play a role. What is clear is that the view that over-bidding in the second price
auction is a universally pervasive phenomenon can be rejected.
In addition to these studies focused on the second price and English auc-
tions, other incentive compatible mechanisms have been investigated as well.
Table 2.3 provides a list of studies that have tested incentive compatible mech-
anisms in induced value studies. Irwin et al. (1998) was the first to study the
BDM mechanism in an induced value setting in both willingness-to-pay and
willingness-to-accept settings. They found that, after five practice rounds, 62%
of willingness to pay bids and 67% of willingness to accept bids were approxi-
mately equal to true value. They could not reject the hypothesis that bids equaled
induced values for either type of valuation. In the willingness-to-pay treatment,
they found that accuracy improved with learning: individuals bid $0.03 closer
to true value, on average, each additional round. Noussair et al. (2004) found
that bids in the BDM were significantly different than true values in all five
rounds of their study; however, while the average bid was 40% less than true
values in round 1, it was only 6% less than true values by round 5.
Shogren et al. (2001a), Parkhurst et al. (2004), and Lusk and Rousu (2006)
investigated the performance of the random nth price auction. The general
results from these studies are that the random nth price auction is demand
revealing in aggregate, but that it performs better for off-margin subjects (i.e.,
those with relatively low valuations) than for on-margin subjects (i.e., those
with relatively high valuations). These studies also showed the second price
auction tended to be demand revealing all along the demand curve, but that it
performs relatively better for high-value individuals. Lusk and Rousu (2006)
also investigated the random nth price auction in addition to the second price
auction and the BDM mechanism. Although they could not reject the hypothesis
of truthful bidding with the BDM, they found that the second price and random
nth price auctions were more accurate at truthfully revealing values than the
BDM.
In summary, performance of incentive compatible mechanisms in induced
value studies reveals equivocal support for demand revealing nature of the
mechanisms. In the few studies that have investigated the English auction,
its performance has been nothing short of stellar; it generates almost 100%
Table 2.3. Results of studies testing incentive compatible mechanisms in induced value experiments
a MP indicates analyses that focused on market prices, IB indicates analyses that focused on individual bids
Incentive compatible auctions: theory and evidence 33
efficiency and generates prices very close to the predicted outcome. The perfor-
mance of the second price auction is mixed. Some studies suggest over-bidding
to be a problem with the mechanism, but others do not observe such behavior.
Evidence exists to suggest that repeated experience with the mechanism helps
improve performance. The BDM mechanism has only received scant attention,
but evidence suggests it is reasonably demand revealing after a period of learn-
ing, though perhaps less so than the second price or random nth price auctions.
The random nth price auction appears to perform well for relatively low-value
individuals, but does not always accurately reveal relatively high bids.
One interpretation of these results is that the mechanisms are too inaccu-
rate to be used to elicit homegrown values that are the focus of this book.
We are more optimistic. Participants in the induced value studies identified in
Table 2.3 were not informed of the dominant strategy in these auctions (Shogren
et al. (2001a) is one exception). Participants either had to infer this strategy from
raw intuition, reasoning, or from trial and error. Considering that many of the
studies in Table 2.3 only used five to 10 bidding rounds, any feedback from trial
and error is weak. What induced values studies have shown is that incentive com-
patible mechanisms can reasonably approximate truthful bidding (better than
ad-hoc or random bidding strategies) even when it is likely that experimental
participants do not explicitly realize the dominant strategy of truthful bidding.
What one need recognize is that when incentive compatible auctions are used
to elicit homegrown values, people are explicitly told the weakly dominant
strategy and are provided with reasoning as to why they should follow it when
bidding. Although some experimentation is likely to take place in experimental
auctions, it is hard to imagine that people would regularly act against their own
self-interest and not offer bids equal to their values.3 This argument is best
summarized by Kagel (1995 p. 513), “. . . there is little doubt that if presented
with Vickrey’s argument [of incentive compatibility], or clear enough examples
embodying that argument, subjects would follow the dominant strategy.”
3 In the context of a bargaining game, Lusk and Hudson (2004) showed that explicitly revealing the
theoretic Nash equilibrium to players served to push outcomes closer to the predicted outcome
compared to when subjects had to act using their own reasoning and intuition.
3 Value theory
3.1 Introduction
For auction markets to be useful in understanding demand for new products,
services, or technologies, we assume people have an economic value, vi , for
the good or service up for auction. This value can be positive, negative, or zero,
but the assumption is that a value does exist. We now discuss exactly what we
mean by the idea of economic value and how it is derived from a utility theoretic
framework.
This chapter discusses the economic concepts of willingness to pay (WTP)
and willingness to accept (WTA), which are the values that are elicited in an
auction. The chapter lays the foundation for how to interpret a person’s value
using the benchmark model of rational individual decision making characterized
in neoclassical economics.1 The chapter first considers how to derive WTP
and WTA for the case when the quality of a good is known with certainty.
Valuation measures are then derived for the case of uncertainty, where a person
attaches probabilities to whether a good possesses particular levels of quality.
For uncertain quality, we start by discussing the expected utility model, and
then we consider a few non-expected utility models. The last part of this chapter
considers how to derive WTP and WTA in a dynamic setting when people have
the ability to delay their decision to buy or sell. Later in the book, in Chapter
9, we extend our discussion to include some of the findings from behavioral
economics and psychology.
34
Value theory 35
to pay (WTP) to purchase the good or his willingness to accept (WTA) com-
pensation to sell the good. The appropriate value measure to use in an auction
depends on whether the person owns the perceived property rights to the good. If
a person does not own the rights to the good, WTP is the appropriate measure of
value. If a person owns the rights, WTA is the appropriate measure of value. The
neoclassical model suggests WTP and WTA should be approximately equiva-
lent if (1) the value of the good is small relative to income, (2) close substitutes
exist for the good, and (3) there is no uncertainty about a person’s preference
for the good (Haneman, 1991).
level U, given market prices and q. The estimated value the consumer places
on the change in the good’s quality from q0 to q1 is:
WTP = m(P, U0 , q0 ) − m(P, U0 , q1 ).
All else being equal, it takes a higher level of expenditure to achieve utility
level U0 if a person is restricted to consume low quality, q0 , instead of high
quality, q1 . Viewed in this way, we can see that WTP is the difference in the
level of expenditure with and without high quality required to generate a fixed
utility level, U0 . There is no difference conceptually or empirically whether
WTP is viewed as a utility maximization problem or expenditure minimization
problem.
An equivalent way to conceptualize WTP is to assume a consumer chooses
both the level of quality, q, and the consumption of market goods, X. Here q
is endogenous. Demand functions, Xm (P, g, y) and qm (P, g, y), result from the
utility maximization problem, where g is the price of q. The dual to this problem
is to minimize expenditures on X and q subject to a given level of utility. The
resulting Hicksian demand for quality is qh (P, g, U). This inverse compensated
demand function for quality, g(qh , P, U) can be viewed as the WTP curve. The
function identifies the price a person will pay for a given level of quality, q,
given
q1 levels of P and U. The WTP for a change in quality from q0 to q1 is
q0 g(qh , P, U0 )dq. Viewing WTP from this perspective makes clear that WTP
for quality changes in one good depends on market prices, P, of market goods.
Our discussion has focused on the final consumer. But researchers are often
interested in estimating a firm’s WTP for a new product or service. Although
WTP is almost always discussed within the context of consumer utility max-
imization, this concept can also be extended to producers. Consider a firm’s
profit maximization decision subject to a given production function. The pro-
ducer chooses the level of inputs, W, to use, but the level of one input, q, is fixed
exogenously. Here q can be thought of as the level of some service provided,
a new technology, or the quality of some input. Given a vector of input prices,
R, and a vector of output prices, P, the firm chooses the optimal level of inputs,
which yields the indirect restricted profit function, π (P, R, q). This indirect
profit function shows the maximum profit that can be obtained given input and
output prices, and a fixed level of q. Now assume the firm considers improving
the quality of an existing input from q0 to q1 . The WTP, or shadow price,
for the change is: WTP = π (P, R, q1 ) − π (P, R, q0 ). Here WTP represents the
maximum amount of profit a firm would be willing to forgo to have q1 rather
than q0 .
how much compensation they would accept for a reduction in the quality of
a good they own from q1 to q0 . Here WTA is the compensation required to
make a person indifferent to the reduction in quality and the status quo. From
the utility maximization perspective, a consumer maximizes, u(X, q), subject
to the budget constraint, XP = y, which yields the indirect utility function
v(P, y, q). The value the consumer places on the quality degradation is given by
the magnitude of WTA such that the following equality holds:
Thus, WTA is the amount of money, that when added to income, makes a
person indifferent to having q1 or q0 . From the perspective of the expenditure
minimization problem, WTA is given by:
m(P, U1 , q0 ) − m(P, U1 , q1 ),
where m(P, U, q) is the indirect expenditure function. One can also derive a
WTA function from a utility maximization problem in which q is endogenous.
Here the inverse demand for q quality is g(qh , P, U); and the WTA for a change
in quality from q1 to q0 is q01 g(qh , P, U1 )dq.
Despite knowledge of the utility derived from high and low quality, the person
does not know with certainty whether a good will be of low or high quality
until after consuming the product. Assume the product is of high quality with
probability p and low quality with probability (1 – p). Expected utility from
consuming the good is:
EU = pU(y, q1 ) + (1 − p)U(y, q0 ).
Now suppose a person is offered the opportunity to choose between purchasing
a good with the uncertain outcome given above and a new product of high
quality with certainty, producing utility of U(y, q1 ) with probability one. The
most a person would be willing to pay to obtain the new good is given by the
ex ante WTP that produces the following equality:
U(y − WTP, q1 ) = pU(y, q1 ) + (1 − p)U(y,q0 ).
WTP is amount of money that makes a person indifferent between having the
high quality good for sure and playing the gamble of consuming the risky good.
If the risk reduction did not completely eliminate the chance of a bad outcome,
then the most a person would be willing to pay for an increase in the probability
of a good outcome from p0 to p1 is:
p1 U(y − WTP, q1 ) + (1 − p1 )U(y − WTP, q0 )
= p0 U(y, q1 ) + (1 − p0 )U(y, q0 ).
Now, consider the case in which a person owns a high quality prod-
uct, in which they derive U(y, q1 ). Suppose the person is asked to sell this
product for one with uncertain quality, yielding expected utility pU(y, q1 ) +
(1 − p)U(y, q0 ). The minimum compensation one will accept to give up the
certain product for the risky one is:
U(y, q1 ) = pU(y + WTA, q1 ) + (1 − p)U(y + WTA, q0 ).
WTA is the level of compensation that makes a person indifferent between
the high quality product with certainty without WTA and the gamble of a low
quality product with the extra income WTA. If the person is offered the chance
to be paid to decrease the probability of a good outcome from p1 to p0 , the
minimum compensation required to accept this change is:
p1 U(y, q1 ) + (1 − p1 )U(y, q0 )
= p0 U(y + WTA, q1 ) + (1 − p0 )U(y + WTA, q0 ).
In the preceeding, it was assumed that when provided with new information
on the probability of a good outcome, people took the information at face
value; however, in reality it is not always true. The probabilities of good and
bad states are subjective and people use various updating rules to incorporate
Value theory 39
new information. To illustrate, consider again the expected utility obtained from
consuming a product:
EU = p̃U(y, q1 ) + (1 − p̃)U(y, q0 ),
where p̃ represents prior beliefs of a good outcome, which is a function of
all accumulated information obtained by the decision maker. Suppose a new
“signal” is received concerning the probability of observing a good outcome:
p . Examples include public information from the government, firm advertising,
word of mouth, or previous experience consuming the product. If a Bayesian
perspective is taken when considering the new information, then prior beliefs
will be updated based on this new information. Following Viscusi (1989) and
Hayes et al. (1995) and assuming a general beta distribution, a person’s belief
in the likelihood
of a good outcome after receiving the new information is
p̂ = α p̃+βp
α+β
, where α is the weight assigned to prior beliefs and β is the weight
put on the new information. Substituting this expression into the expected utility
formula yields:
αp̃ + βp αp̃ + βp
EU = U(y, q1 ) + 1 − U(y, q0 ).
α+β α+β
The person’s WTP to obtain q1 for sure is now written as:
αp̃ + βp αp̃ + βp
U(y − WTP, q1 ) = U(y, q1 )+ 1 − U(y, q0 ).
α+β α+β
Totally differentiating this expression with respect to WTP and p , and rear-
ranging terms indicates that:
dWTP β U(y, q0 ) − U(y, q1 )
= ≤ 0. (3.1)
dp α+β dU (y, q1 )/dy
Equation (3.1) shows that WTP to obtain q1 is decreasing in p . If a person
learns there is a higher likelihood of receiving the high quality outcome from
consuming the good, they will pay less to obtain a good that is high quality with
certainty. But as equation (3.1) shows, WTP is conditioned by how the person
weighs the new information relative to their prior information. For example, if
β = 0 the person ignores the new information and WTP is unchanged. Inci-
dentally, the right-hand side of the equality in equation (3.1) is the same for
WTA. WTA to consume the good with an uncertain outcome falls once one
learns there is a greater likelihood q1 will be obtained and the magnitude of the
change is conditional on the relative magnitudes of α and β.
It is also useful to consider how people value changes in a continuous set
of risky outcomes as opposed to the simple dichotomous good/bad outcomes
just discussed. Suppose a person evaluates the desirability of consuming a new
good, and the outcome of this consumption is risky. Let Z̃ be a random variable
40 Experimental Auctions
do you think is the chance of becoming ill from (pathogen name), given that
you eat an average amount of typical products in the United States over a
year? Answer: chance out of 1 million people.” Stating an exact proba-
bility, especially for a low-probability event, can be a difficult task for people.
An alternative approach is to use psychometric scales. For example, Pennings
et al. (2002) measured risk perceptions by asking people questions like, “I think
eating beef is risky” on a scale of 1 (strongly disagree) to 9 (strongly agree).
Lusk and Coble (2005) took a similar approach to measure risk perceptions by
asking questions like, “The side-effects from genetic modification in food pro-
duction are largely unknown” on a scale of 1 (strongly disagree) to 9 (strongly
agree).
Measuring risk preferences can be more involved. One approach is to use
economic experiments to estimate Pratt-Arrow coefficients of risk aversion. For
example, Holt and Laury (2002) introduced an approach that has been used in
many subsequent studies (e.g., Lusk and Coble, 2005). Their approach asks
people to make ten choices between two lotteries that have uncertain monetary
outcomes. Choices are used to identify people’s Pratt-Arrow coefficients of
relative risk aversion. Another approach might involve asking people to state
their certainty equivalent for monetary lotteries as in Becker et al. (1964),
which can be used to construct a utility function for each person. A less precise,
but easy-to-implement approach is to use psychometric scales to estimate risk
preferences. For example, Pennings et al. (2002) measured risk preferences by
asking people questions like, “I am willing to accept the risk of eating beef” on
a scale of 1 (strongly disagree) to 9 (strongly agree).
us little about how to deal with such low-probability threats. Numerous exper-
imental studies show that people commonly overestimate the chance that they
will suffer from a risk with low odds and high damage. People use heuristics,
or rules of thumb, to deal with risk (e.g., see Kahneman and Tversky, 2000;
Tversky and Fox, 1995). People choose to think about probabilities and out-
comes separately or in sequence, rather than in combination as predicted by
expected utility theory. Economists and psychologists continue to test whether
notions of expected utility or alternative models of behavior better explain how
people make choices under risk and uncertainty.
Two recent models that attempt to capture this non-expected utility behav-
ior include rank-dependent expected utility theory (Quiggin) and cumulative
prospect theory (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992).
In these models, a person does not necessarily weight probabilities linearly. A
probability weighting function, w( p), is used to transform probabilities into
weights. The probability weighting function is a strictly increasing function
with bounds at zero and one.
First, consider the rank-dependent expected utility model, which is:
n
VRD−EUT = πi U (y, xi ), (3.5)
i=1
where πi = w nk=i pk − w nk=i+1 pk for in − 1, πn = w( pn ), and
x1 ≤ . . . ≤ xn . Various forms have been proposed for the weighting function:
two common forms are w(z) = z γ and w(z) = z γ /[z γ + (1 − z)γ ]1/γ , where γ
is a parameter indicating the shape of the weighting function. Equation (3.5)
models behavior under risk as ranking the outcomes in order of preference and
then distorting the decumulative probabilities through a probability weighting
function. To illustrate, consider a lottery with a p1 chance of x1 and p2 chance
of x2 (where x1 < x2 ), then the rank-dependent expected utility of the lottery
is (1 − w( p2 ))U (y, x1 ) + w( p2 )U (y, x2 ). Rank-dependent theory collapses to
expected utility theory if w( p) = p or if γ = 1. People’s valuations under
rank-dependent expected utility theory are similar to that in expected utility
theory for a 100% reduction in risk. WTP to obtain an n-outcome lottery is:
n
U (y − WTP) = πi U (y, xi ).
i=1
functions for gains and losses relative to some reference point (Tversky and
Kahneman 1992). For the sake of convenience, assume the reference point is
zero and x1 ≤ . . . ≤ xr ≤ 0 ≤ xr +1 . . . ≤ xn . The utility of a lottery in cumu-
lative prospect theory is:
r
n
VCPT = πi− U − (xi ) + π+ +
j U (x j ), (3.6)
i=1 j=1
where the decision weights for losses, πi− , and gains, πi+ , are separately defined
in a manner similar to that above and where U − (x) and U + (x) are the utility
functions for losses and gains, respectively. Cumulative prospect theory reduces
to rank-dependent expected utility if w− ( p) = 1 − w+ ( p) for all p and U − (x) =
U + (x) for all x. In cumulative prospect theory, valuation measures, WTP and
WTA, depend on the initial reference point and whether a lottery is perceived
as a gain, in which the formulas w+ ( p) and U + (x) apply or as a loss in which
the formulas w− ( p) and U − (x) apply.
3.5 Summary
This chapter provided background on theoretical models that are useful in inter-
preting the values elicited in experimental auctions and in identifying determi-
nants affecting valuations. We began by discussing the benchmark neoclassical
Value theory 45
4.1 Introduction
In this chapter, we address some broad issues worth considering before start-
ing off on the path of choosing specific auction parameters. Perhaps the most
important initial step in any project is to define the study objectives. What
are the goals of the analysis? What are the testable hypotheses? Who is the
intended audience? Answering these questions will necessarily dictate many of
the resulting design choices. A study that aims to test a particular behavioral
or economic theory might require a significantly different design than one that
aims to estimate the welfare effects of a new technology or to inform new prod-
uct design. For example, a common complaint of experimental studies is that
students are frequently used as subjects. The veracity of this criticism, how-
ever, depends on the nature and type of study. A theory is a generalization that
should hold for everyone, including students. As stated by Noussair, Plott, and
Riezman (1995, p. 462):
Since the world’s international economies are vastly more complicated than the
economies created for this study, of what relevance are laboratory-generated data? The
answer is that laboratory experiments are not attempts to simulate field situations, as
the question of the skeptic seems to presume. Laboratory research deals with the gen-
eral theories and the general principles that are supposed to apply to all economies,
the economies found in the field as well as those created in the laboratory. The lab-
oratory economies are very simple and are special cases of the broad class of (often
complex) economies to which the general theories are supposed to be of relevance. If
the general theory does not work successfully to explain behavior in the simple and
special cases of the laboratory, it is not general. When a model is found not working,
opportunities exist to modify the theory to account for the data or to reject the theory.
Thus, the laboratory provides an arena in which competing notions and theories about
the nature of human (and market) capacities can be joined with the data.
46
Conducting experimental auctions: some preliminaries 47
Although it might appear trivially obvious that one should define a study’s
objectives, this critical step can be overlooked. In addition to defining the objec-
tives, several other preliminaries relating to implementing experimental auc-
tions should be considered. This chapter considers the issues of experimental
design, sample size, and study setting. Experimental design primarily concerns
itself with collecting data such that the effect one is interested in measuring is
actually what is measured and is not confounded with other effects. Another
key purpose of experimental design is to collect data in such a way as to mini-
mize the standard error of estimated effects. In addition to investigating how to
collect data via experimental design, this chapter also discusses how many par-
ticipants should be enrolled in a study. Here again, determination of sample size
depends on the purpose of the study. If the purpose is to estimate a population
statistic using a random sample of individuals from the population, the idea
is to minimize the sampling error. If the purpose is to test for means between
two treatments, the idea is to maximize the power of the test. In both cases, a
key factor influencing the appropriate number of participants is the anticipated
variability in valuations across people; the more diverse the opinion, the larger
the sample size required. Finally, this chapter considers where an auction should
be conducted; the laboratory or the field.
Product information
Small A B
Large C D
Table 4.1 shows a full factorial design, where all possible combinations of
variables and levels of variables are considered. In this case, the full factorial
consists of four distinct treatments (A, B, C, and D). This design allows one
to test the linear effects of information and package size on auction bids (e.g.,
the effect of moving from high to low information and the effect of moving
from small to large package size), and any interaction effects that might exist
(e.g., the effect of information on auction bids conditional on package size).
The point of experimental design is to identify the effects of interest. Suppose
data were only obtained from treatments A and D; if so, it would be impossible
to determine whether a change in auction bids from treatment A to D was
due to package size or information. The treatment effects are confounded. If
one only wanted to conduct two treatments and avoid confounding the design,
one could carry out treatments A and C while holding product information at
low. Such an experiment would identify the effect of package size on bids.
This example of confounded treatment effects is trivially obvious and few
practitioners would fall prey to such a mistake. But real world experiments are
seldom as simple. There are many variables of interest in addition to product
information and package size. To illustrate, consider the addition of another
explicit treatment variable, product color, which is varied between two levels:
red and blue. Now there are 23 = 8 possible treatments in the full factorial design
as shown in Table 4.2. Suppose resources are too scarce to collect data from
all eight treatments. How can one obtain a smaller number of treatments and
still estimate the effects of interest? One method involves creating fractional
factorial designs that uses portions of the treatments from the full factorial by
making assumptions about higher-order effects.
Probably the most common fractional factorial is a main-effects only design –
a design where all linear (or main) effects are separately identifiable (Louviere
et al., 2000). Table 4.2 illustrates a full factorial and a main-effects design taken
there from. One can identify the linear effects of information, package size, and
product color by only collecting data associated with treatments F2 (high, small,
blue), F3 (high, large, red), F5 (low, small, red), and F8 (low, large, blue) from
the full factorial. Data only need be collected on four treatments instead of eight
so long as one assumes no interaction effects exist. But how is this possible? The
Conducting experimental auctions: some preliminaries 49
Variables
Variables
Information
Treatment (x1) Size (x2) Color (x3) x1*x2 x1*x3 x2*x3 x1*x2*x3
Orthogonal fraction
profiles in design Correlation matrix (X X)−1
1
x1 x2 x3 x1 x2 x3 x1 x2 x3 1(X X)−1
F2 −1 −1 1 x1 1 0.25 0 0
F3 −1 1 −1 x2 0 1 0 0.25 0 64
F5 1 −1 −1 x3 0 0 1 0 0 0.25
F8 1 1 1
Non-orthogonal
fraction profiles in
design Correlation matrix (X X)−1
x1 x2 x3 x1 x2 x3 x1 x2 x3
individual. One must determine how people should be assigned to the treatments
and how many replications (e.g., how many subjects) to conduct (we discuss
the number of subjects in Section 4.3). Instead of a person, the experimental
unit may be an auction session if, for example, the price generated by the
auction is the statistic of interest. If so, the question is how many experimental
sessions should be conducted for each treatment. In either case, it is important
to realize there can be session-specific effects which implicitly serve as separate
treatment variables. For example, in Table 4.1, if data were obtained from only
one session in each of treatments A, B, C, and D, one could not be certain that the
difference in auction bids between sessions A and C was truly due to package
size or whether it was due to session-specific effects. An easy-to-implement and
powerful approach to assign experimental units to treatments is randomization –
each unit is randomly assigned to a treatment.
In addition to explicit design variables, there are other extraneous variables
that might influence auction bids. For example, the time of the day or the day
of the week that the experiment is conducted has been shown to influence bids
(see Hoffman et al., 1993). The influence of such factors might result from
the dynamics of individual preference (e.g., people are hungrier closer to meal
time) or to differences across people that result in selection effects (e.g., only
people without a job will show up at sessions during the day). One way to hold
constant such factors is to conduct all experimental treatments at the same time
and day with people randomly assigned to a treatment. All too often university
professors are apt to assign all students in a class to a single treatment or perhaps
to assign all early arrivals to an experiment to a single treatment. One can see
how such practices can lead to confounding influences if, for example, one class
is made up of seniors taught in the evening and the other class is made up of
juniors taught in the morning or if early arrivals have different attitudes and
preferences than late arrivals. Logistically, it may not be feasible to carry out
all treatments at the same time, and if so, one might consider making time, day,
classification, etc. explicit treatment variables if the effects are anticipated to
be large.
There will, of course, be some variables that are difficult or impossible to hold
constant across subjects in an experiment. Many times these are subject-specific
variables such as knowledge or income; we cannot reasonably vary a person’s
income from $20,000 to $100,000 and investigate the effect of this change on
auction bids. There are a couple of approaches that might be followed to control
for such variables. First, one can use a within-subject design, where a single
person is assigned to more than one treatment. For example, in Table 4.1, a
person might first participate in treatment C and then in treatment D. Because
the individuals’ age, income, gender, etc. are unchanged (i.e., held constant), the
difference in behavior should be due to the treatment effect of interest, product
information in this case. A word of caution: within-subject designs can create
order effects due to learning or fatigue. A change in behavior from treatment
Conducting experimental auctions: some preliminaries 53
C to D may be because a person was tired of putting cognitive effort into their
bidding decisions or because they learned more about the mechanics of the
auction. A common method to alleviate such concerns is to use what is called
an ABA design. People first participate in treatment A, then treatment B, and
then treatment A again. Differences in behavior between the two A treatments
can be investigated to determine whether learning or fatigue is a problem.
Further, ABA experiments can be used in combination with their mirror image,
BAB experiments, to average out the order effects. A second word of caution in
within-subject experiments is demand reduction. If a subject makes a purchase
in treatment A, they will move down their demand curve and will bid less for a
second unit in treatment B. Chapter 5 provides more discussion on the effects
of demand reduction and provides suggestions for dealing with the problem.
The most straightforward way to handle the potential problem is to only make
one of the treatments binding, by randomly picking one of the treatments after
the experiment is complete.
A final method to control for other confounding variables is the one familiar
to most economists: randomly recruit a large sample of participants and use
ex-post regression analysis to determine the ceteris paribus treatment effects.
Data can be collected from people after the experiment on income, age, educa-
tion, etc., and regression equations can be estimated in which the auction bids
are regressed against these demographic variables and dummy variables for
treatment effects. The estimated coefficients on the treatment dummy variables
indicate the effect of the experimental treatment effect holding constant the
effect of demographics. More information on regression analysis with auction
bids is included in Chapter 6.
In our motivating example, we assumed all other variables besides prod-
uct information, package size, and color were held constant at some fixed level.
Subsequent discussion mentioned a variety of ways to practice control in exper-
iments. Control is a multifaceted issue. For example, in Chapter 3 on value
theory, we saw that auction bids can depend on the price of substitute goods.
If subjects are uninformed about the price of an important substitute, much of
the variation in auction bids may be due to heterogeneity in individuals’ per-
ceptions about the price of the substitute and not any pertinent treatment effect,
per se (e.g., see the discussion in Harrison et al., 2004). In such a case, the
researcher has lost control over their experiment in the sense that a potentially
sizable portion of the variation in auction bids is now unexplainable. A straight-
forward way to handle this problem is to inform subjects of the price of the
substitute good outside the laboratory such that all participants have a common
expectation or perhaps to offer the substitute good for sale at a fixed price in the
experiment.
Other factors also have the potential to affect auction bids and the key is to
determine the factors most salient and hold them fixed at some constant level
in the experiment. Unless the effects of these variables on auction bids interact
54 Experimental Auctions
with the effects of the key treatment variables of interest, it is not important at
what level the variables are held (e.g., high or low, red or yellow, yes or no, etc.),
simply that they are held constant. In many ways, the choice of what variables to
explicitly consider in a design, which to hold constant, and which to vary, is as
much an art as a science. Theory can provide a guide regarding factors expected
to influence bids, but even here guidance is limited. For example, it was just
argued that, based on theory, the price of substitute goods is expected to influ-
ence auction bids; however, many questions remain. Which substitute goods
should be considered? How many? In this section between-subject designs,
where the same person takes part in several treatments, were suggested as a
way to hold constant many factors. Later in the book we discuss how endowing
subjects with a good and having them bid to “upgrade” to a substitute can also
hold constant many potentially compounding factors. Sill, there are likely to be
significant differences across researchers as to the importance of a particular
factor in a particular study.
One researcher’s gain of control can be another’s loss of context. Some
researchers argue that it is these confounding factors (e.g., uncontrolled body
language sent in face-to-face bargaining) that provide the rich economic context
that motivates real-world behavior. They point out that striving for complete
control is self-defeating because it creates an economic environment that is too
sterile, too abstract, too unreal, with no parallel in the real world, and generates
subjects who are motivated by salient payoffs, but unmotivated by the context in
which these payoffs are earned. The inspiration is that economic theory should
be robust enough to explain actions in real world social interactions that have
copious contextual cues that occur simultaneously. In response, advocates of
more control point out that too much context makes the experiment too much like
an open-ended case study or descriptive naturalist observations. Experiments
based on too many confounding factors and misidentified motivation yield
no universal patterns based on first principles, which creates little ability to
generalize beyond the specific case. One might even argue this approach is non-
experimental. The recommendation is that when examining such contextual
issues one should note whether the method permits the conclusions that the
author has made. A causal inference may have been made when the design
does not allow such inference.
This section outlined many basics of experimental design. There are many
good reference books on the topic that can provide more depth and insight. For
examples, see Kirk (1994) for a detailed text on experimental design from the
perspective of a psychologist, Montgomery (2000) for a general text written
by an engineer, Friedman and Sunder (1994) for a discussion on experimen-
tal design in relation to experimental economics, and Louviere, Hensher, and
Swait (2000) for discussions on experimental design in relation to stated choice
surveys.
Conducting experimental auctions: some preliminaries 55
2(z α + z β )2 σ 2
.
2
56 Experimental Auctions
5 10 11 12
10 16 18 19
15 22 24 26
20 27 30 32
25 33 36 38
30 39 42 44
35 44 48 50
40 50 53 56
45 55 59 62
50 60 65 68
55 66 70 74
60 71 76 80
65 77 81 85
70 82 87 91
75 87 92 97
80 93 98 102
85 98 103 108
90 103 109 114
95 109 114 119
100 114 120 125
Shogren, 1998; Shogren et al., 1999; List, 2001; Lusk et al., 2001a; Lusk and
Fox, 2003; see the review by Harrison and List, 2004).1
There are advantages and disadvantages to field and laboratory experiments.
The primary advantage of conducting experiments in a laboratory setting is
more control. In a laboratory, an experiment can be constructed in such a way
as to minimize confounding factors such that the effects of interest can be
isolated. When dealing with the elicitation of homegrown values, our focus,
one must recognize that while laboratory permits control over many aspects of
a design, one has to contend with “outside” market influences. For example, a
rational person would never bid more for a good in a laboratory auction than
the price of the good in the market outside the lab (plus some transaction costs).
The laboratory, however, gives a researcher control over both the content and
sequencing of price and non-price information provided to subjects.
Moving the valuation setting to more familiar territory such as a grocery store
or a mall might be advantageous for a number of reasons. First, subjects self-
select into the market, and as such, sample-selection biases are of less concern
as the population of interest is directly intercepted. Second, in a field setting,
subjects are able to bring all their learned knowledge to bear on the task at hand.
To the extent that individuals develop heuristics to make purchase decisions in
the marketplace, conducting an experiment in a field setting allows individuals
to use such heuristics. Many times, field experiments provide interesting tests
of the effect of market experience on behavior. The effect of market experience
on behavior is non-trivial and has important economic consequences (e.g., see
Crocker et al., 1998 and List, 2003).
Finally, in many instances field valuation can reduce the costs of experimental
work. This cost reduction is potentially achievable because compensatory fees
must often be paid to subjects to attend a laboratory session, but such fees
are unnecessary in the field. Because the magnitude of compensatory fees has
1 Harrison and List (2004) came up with a nomenclature to categorize experiments done with
the general population: a) artefactual field experiments that take place in a laboratory setting,
but use “real” people as subjects as opposed to a convenience sample of students, b) framed
field experiments are similar to laboratory experiments except they occur within the field context
(in either the commodity, task, or information set that the subjects use), and c) natural field
experiments that are similar to framed field experiments except that subjects do not know they
are taking part in an experiment. To illustrate, consider an experiment aimed at determining the
value a person places on a new food product. A typical laboratory experiment would involve
student subjects bidding on the new good in a classroom or laboratory. With an artefactual field
experiment, one would randomly recruit people representative of the market of interest and
would elicit auction bids for the new good in a classroom or laboratory. A framed experiment
would involve the experimenter moving to the place where individuals actually make the purchase
decision (e.g., grocery store, mall, internet, etc.). Then subjects would bid in an auction for the
new good in the environment that matches that where subjects would consider purchasing such a
good. A natural field experiment would be similar to the framed experiment except that subjects
would not know they were participating in a research project. An example of a natural field
experiment might be the auction of a good on e-bay.
Conducting experimental auctions: some preliminaries 59
been shown to influence bids (e.g., Rutström, 1998), field valuation offers the
potential to omit this bias. A variety of methods have been used to attract
participants to participate in a field auction including offering a free unit of
a “traditional” good and eliciting bids to exchange the endowed good for a
“new” counterpart (e.g., Lusk et al., 2001a) and offering coupons to the retail
establishment in which the elicitation task is taking place (e.g., Lusk et al.,
2006a). In a natural experiment, no inducement is needed; subjects select into
the market because they are interested in purchasing the good.
Lusk and Fox (2003) directly tested whether auction bids elicited in a labo-
ratory setting were equal to those elicited in a retail setting. After controlling
for zero bidders, they found that bids were significantly higher in the store
setting than in the lab; about 30% higher after controlling for other exoge-
nous factors. They suggest this difference may be due to value uncertainty and
impatience. Zhao and Kling (2001) illustrate that in a dynamic context that
willingness-to-pay should increase when consumers are more certain about a
good’s value, more impatient about consuming the good, and expect less infor-
mation can be gathered about a good’s value in the future. In the retail setting,
individuals enter the store ready to make a purchase and are more impatient
than someone sitting in a laboratory. Further, the store setting has more infor-
mation about substitute prices for goods, which should make subjects more
certain about a good’s relative value and decrease the chance that more infor-
mation will be gathered later, which would serve to increase willingness-to-
pay.
Another key study was conducted by Shogren et al. (1999), who compared
the similarity of lab valuation choices to field retail store choices for risk reduc-
tion. They also compared choices made in hypothetical surveys. All subjects
came from the same small college town and made choices between typical
chicken breasts versus chicken breasts irradiated to reduce the risk of food-
borne pathogens. The results from both the survey and experimental market
(note: the experimental market was not an auction market, it consisted of partic-
ipants making non-hypothetical choices) suggested significantly higher levels
of acceptability of irradiated chicken than in the retail setting at an equal or
discounted price for irradiated chicken. However, consumer choices were more
similar across market settings when irradiated chicken was sold at a premium
over regular chicken. They observed that in a mail survey and a lab experiment,
both of which included information about irradiation, 80% of participants pre-
ferred irradiated to non-irradiated chicken breasts when they were offered at the
same price; however, only about 45% of shoppers in the retail setting bought
irradiated chicken when it was priced the same as non-irradiated. In contrast,
when the irradiated product was offered at a higher price, the survey and exper-
imental results predicted market share in a subsequent retail trial remarkably
well. In all three settings (survey, experimental market, and store), about 33%
60 Experimental Auctions
of people bought the irradiated chicken when it was priced at a 10% premium
over regular chicken.
While differences in choices across institutions were observed in Shogren
et al. (1999), each of the three decision settings involved unique features
and incentives that were absent from the other two settings. Retail purchases
involved payment of real money for real products within an environment in
which thousands of products competed for the consumer’s dollar. Any attempt
to collect consumer information in a retail setting was liable to interfere with
the realism of the market itself. The goal of the retail setting was to establish
the most realistic baseline possible against which one could judge the lab or
survey. In contrast, while the experiments also required exchange of real money
for real goods, the participants knew they were being monitored and the range of
alternative purchases was limited. The survey involved hypothetical decisions,
information about irradiation, and people knew that they would not be held
accountable for choices they made. Perfectly simulating a retail experience in
the lab or a survey so as to control every nuance is unattainable. Rather the goal
should be to compare lab and survey choices relative to a real world baseline.
The research program that emerges for future work is to explore how compre-
hensive the lab or survey environments must be to come closer to replicate an
actual market outcome.
An important consideration in the decision of whether to carry out valuation
in the lab or field regards how consumers view their participation in the exper-
iment. Levitt and List (2005) proposed a model of individual decision making
to explain how responses from laboratory experiments might diverge from the
“real world.” Their model assumes people derive utility from wealth and moral-
istic concerns. Levitt and List (2005) posit that the weight a person attaches
to the latter of these concerns depends on the extent to which one’s actions
are scrutinized or “watched.” This idea revisits the classic Hawthorne effect.
Experimenters examining job performance made changes in the workplace of
the Hawthorne plant of the Western Electric Company in Chicago that made
people feel important and thus improved their performance. In any survey situ-
ation, people know their actions are being scrutinized, and as such, they could
have incentive to provide more pro-social and moralistic answers as compared
to what they might do in a naturally occurring marketplace.
The contribution of studies such as Lusk et al. (2001a), List (2001), Bohm
(1994), and others, in which the valuation setting was moved to the field, but
where the researchers retained control over certain key design parameters, is that
they represent an effort to bridge the gap between the uncontrolled, yet realistic
retail setting and the highly controlled, yet abstract lab setting. Practitioners
now choose where to conduct their experiments along the continuum ranging
from laboratory auctions with students to natural field auctions with naturally
occurring participants.
Conducting experimental auctions: some preliminaries 61
4.5 Conclusions
This chapter introduced and discussed some preliminary issues associated with
the design of experimental auctions. The chapter covered inter-related issues
associated with control over the experiment (e.g., experimental design), context
(e.g., field versus the laboratory), and the number of participants (e.g., sample
size). The purpose of experimental design is to collect data in a way so as
to identify all the “effects” one is interested in. Experimental design involves
identifying the variables most likely to influence auction bids and combining
the levels of these “important” variables to create treatments. There is no magic
bullet for determining what variables are “important” and should be included
as an experimental control.
We offered up several pieces of advice. Researchers should hold constant as
many factors as possible either through implementing between-subject designs,
randomization, eliciting bids to upgrade from an endowed good, using ex-post
regression analysis, or creating instructions and procedures to create common
expectation among participants. Another key purpose of experimental design
regards the efficiency of estimates. The more efficient an experimental design,
the fewer observations necessary to reject a hypothesis. Lusk and Norwood
(2005) show that, provided an experimental design does not confound two
effects, large sample sizes can compensate for poor experimental design.
In addition to concerns with efficiency, the chapter also discussed sample size
from a sampling standpoint. Because every person in the population does not
participate in an experimental auction, the mean auction bid from a sample of
recruited experimental auction participants will consist of some level sampling
error. This chapter illustrated how to determine the level of sampling error for a
given sample size. The section on sample size also discussed how to determine
the minimum sample size required to test the null hypothesis of equality between
two treatments given assumptions about the power of the statistical test. The
chapter concluded with a discussion regarding the setting (field versus lab) of
the valuation experiment. In general, laboratory auctions permit more control
over the decision environment, but this comes at the loss of context and realism.
When designing experimental auctions practitioners must weigh the importance
of a number of competing factors including control, context, and sample size
and make judgment calls regarding the importance of each factor relevant to
the study objective.
5 Conducting experimental auctions
5.1 Introduction
Researchers have explored and debated many issues on the proper design and
conduct of experimental auctions used to elicit values for lotteries, new goods,
services, and technologies. This chapter examines six essential design issues in
detail – training and practice; endowment of a good versus full bidding; choosing
an auction mechanism; multiple good valuation, demand reduction and field
substitutes; learning and affiliation in repeated bidding rounds and negative
values. For each issue, we discuss the choices available to the practitioner,
outlining the pros and cons that have emerged from theoretical and empirical
literature. The goal of this chapter is to make readers aware of the relevant
issues so that informed and justifiable choices can be made when designing an
experimental auction.
Some obvious “best practices” have emerged that should be followed in con-
ducting auctions. For example, practitioners should initially conduct a quali-
tative study prior to designing an experiment such that they can learn about
individuals’ decision-making processes and explore how people think about
the good and experimental procedures in question. Focus groups in conjunc-
tion with pre-tests of the experiment will help ensure that the study’s objectives
are properly met. Once a design has been settled on, it is critical that subjects
properly understand the auction in which they participate and that every effort
is made to ensure against misperceptions. Aside from these issues, the practi-
tioner has some judgment calls to make. Thinking through these issues when
deciding what choices to make will also introduce a new set of concerns and
challenges into the literature.
62
Conducting experimental auctions 63
take care to ensure that people do not attempt to use the optimal strategies
from those auctions, in which truthful bidding is not necessarily the dominant
strategy.
We believe it is critical that prior to conducting the main auction, people
should: (a) receive training on the incentive compatible mechanism; (b) partic-
ipate in real-money practice rounds with another good; and (c) be assured of
anonymity. Without such safeguards there is little reason to believe that elicited
bids will correspond to individuals’ values. For example, Brown (2005) used a
BDM mechanism to elicit values for several goods without using any practice
rounds. The contribution of Brown’s study was the use of a “verbal protocol,”
in which people say aloud what is going through their mind as they formulate
their bids. Brown (2005) found that most subjects mentioned the item’s value to
others, the expected market price, or the opportunity for making a profit when
formulating their bids. These findings indicate that people have misperceptions
about the incentives in the auctions. Such findings would likely induce some
to question the ability of auctions to elicit true values at all. For example, in
a study comparing bids from second and ninth price auctions, Knetsch, Tang,
and Thaler (2001) concluded (at p. 265), “Contrary to common understanding,
a Vickrey auction may not be demand or value revealing . . .” This conclusion
is premature. Plott and Zeiler (2005) provide evidence that misperceptions play
an important role in individuals’ bidding and that these can be removed with
careful experimental design that includes training and experience.
Without any training or experience, people will probably resort to heuristics
they have learned by participating in other markets, such as buy low and sell
high. They must be brought to the realization that such strategies are not the best
response in incentive compatible auctions. This can be partially accomplished
through good explanations of the properties of mechanisms, but experience with
the mechanism is critical as well. Real-money practice rounds (with a non-focal
good) permit people to learn first hand the perils of bidding sub-optimally.
Prior to conducting an auction, we recommend that nine steps are imple-
mented.
Step 1. Carefully explain the mechanics of the auction. This means creating
written instructions that have been tested on a control group to eliminate any
confusing statements and other misunderstandings. One might find it helpful to
use terms that discourage participants from engaging in behaviors that deviate
from truthful value revelation. For example, using words like “buyer” or “seller”
rather than “winner” might serve to mitigate “auction fever.”
Step 2: Provide concrete numerical examples. To avoid confounding influ-
ences, examples can be contextualized with a fictitious good or different good
than the one of interest. Candy bars can provide a useful reference good both
to use as an example and for practice rounds. A numerical example for the sec-
ond price auction might go something like the following. Suppose there were
64 Experimental Auctions
five people participating in an auction like the one described. Suppose that
these people were bidding to obtain a Snickers bar. Participant #1 bid $0.00
for the Snickers bar, participant #2 bid $0.10, participant #3 bid $0.25, par-
ticipant #4 bid $0.40, and participant #5 bid $0.50. Participant #5 would win
the auction because they bid the highest amount and they would pay $0.40,
which was the second highest bid. Participants #1, #2, #3, and #4 would not
pay anything and would not receive the Snickers bar. Examples might also be
followed with a disclaimer that the monetary values are used for illustrative pur-
poses and are not meant to convey what the goods should or might be worth to
participants.
Step 3: Clearly explain why it is in each person’s best interest to bid truthfully.
Despite the incentives inherent in mechanisms such as the second price auction,
these incentives are probably not obvious to people that have never used the
auction. The incentive structure of most auctions is easily conveyed to people.
A variant on the following would serve to explain the incentives: In this auction,
the best strategy is to bid exactly what it is worth to you to obtain the good.
When you submit your bid, you do not know what the price will be. If you bid
more than the candy bar is worth to you, you may end up having to buy the
good at a price that is more than you really want to pay. By contrast, if you bid
less than the candy bar is really worth to you, you may end up not winning the
auction even though you could have bought the candy bar at a price you were
actually willing to pay. Your best strategy is to bid exactly what each candy bar
is worth to you. Numerical examples, could further illustrate the advantages to
truthful bidding. Suppose your value for a Snickers bar is $0.50. If you decide
to offer a bid less than your value, say $0.40, you may not win the auction
if the market price is $0.45. This means that by under-bidding you have given
up the opportunity to buy a Snickers bar for only $0.45 when it was worth $0.50
to you. By contrast, suppose you decide to bid $0.60, which is more than the
Snickers bar is worth to you, and you end up winning the auction. If the market
price is $0.55, you would have to buy a Snickers bar for $0.05 more than it is
really worth to you.
Step 4: Use a simple quiz to test individuals’ knowledge of the mechanism.
Asking people to answer questions about the mechanics of the auction encour-
ages people to think through the auction mechanism and allows the researcher
to correct any misperceptions that might exist. For example, one could re-pose
a Snickers bar auction scenario and ask people to indicate the winning bidder(s)
and the market price. It might be indicated that Participant #1 bid $0.01 for the
Snickers bar, participant #2 bid $0.17, participant #3 bid $0.05, participant #4
bid $0.52, and participant #5 bid $0.37. Then people can be asked questions
such as: Who would win this auction? How much would the winner pay?
Step 5: Allow questions that pertain to the mechanics of the auction. To
avoid creating confusion between experimental treatments, researchers may
Conducting experimental auctions 65
wish to avoid some types of question. A formal protocol can be developed for
responding to potential questions.
Step 6: Conduct several rounds of real-money practice auctions with a non-
focal good. Although good written and verbal instructions are key, there is
no substitute for actual experience with the mechanisms. People should have
the opportunity to participate in several real-money auctions. A number of
possibilities exist to help ensure that behavior from the practice auctions does
not confound valuations for the good of interest. One possibility is to auction
a good that is unrelated to that which is the primary focus of the investigation.
Again, candy bars may serve as a useful practice good. Experience can be gained
by conducting a series of auctions for different candy bars (e.g., Snickers, Milky
Way, etc.) or by repeating auctions for the same candy bar. Another possibility
is to auction generic lotteries. For example, people could bid to obtain a 50%
chance of winning $1 and a 50% chance of winning $0 (as determined via a
coin flip). Another option is to use an induced value auction to teach people
about the mechanism.
Step 7: Impose anonymity. If people are not sure their identities are confi-
dential, they could submit bids to look good to other participants or to please
the experimenter. A lack of anonymity could cause people to offer bids related
to how others might value the good (e.g., a “fair” market price). The assign-
ment of ID numbers and individualized pay-outs at the end of the experimental
session can be used to impose anonymity. Some experimental labs add an
additional layer of anonymity by using individualized keys and mail boxes in
which people receive their recruitment fee and experimental earnings (minus
the price of any purchased items) and any goods they might have purchased
during the experiment. The practice auctions can be carried out with or without
anonymity. Relaxing the anonymity restriction during some portion of the prac-
tice can improve understanding as people can ask questions during the auction
and the experimental monitor can use bids to further explain the properties of
the auction.
Step 9: Once a person has been trained in the auction mechanism, one can
run the real-money auction for the focal good.
approach and the full bidding approach. The endowment approach has people
bidding on “upgrading/downgrading” from one endowed good to another good.
The full bidding approach has people bidding on both goods simultaneously.
Consider the pros and cons of each approach.
The endowment method involves giving subjects one good, X, and eliciting
bids to exchange X for another good, Y. Bids from the endowment approach
directly reflect the value difference between X and Y. An advantage of the
endowment method is that it mitigates outside-market influences. Suppose good
X is a typical good sold in the market and Y is a new good to be valued.
Consumers can buy X at market price $P outside the experiment, but Y is only
available in the experiment.
Assume a consumer’s true values for X and Y are $VX and $VY . Using the
endowment method, one directly elicits $VY − $VX , which is not confounded
with outside lab opportunities (assuming opportunities for resale of Y in the
field are limited).
The full bidding approach involves subjects bidding simultaneously to obtain
one of two goods, X or Y. To avoid demand reduction effects, only one of the
goods is randomly selected and sold; this determination is made after both
bids have been submitted. Now, if the full bidding method is used with an
incentive compatible auction, a rational consumer’s bid for X is: min($VX ,
$P + c), where c is the perceived transaction cost of purchasing X in the field
rather than in the lab. A consumer will bid their true value, $VX , in the experi-
ment only if it is less than the price for which they can buy the good in a grocery
store or vending machine – the bid for X is censored at the market price plus
some transactions cost. By contrast, because Y is only available in the lab, the
consumer’s bid for Y is unbounded and should be equal to $VY . So, using the
full bidding method, the difference in bids might be $VY − $P − c which is
not the measure of real interest: $VY − $VX .
Corrigan (2005) provides a similar discussion related to option values. If
values are uncertain in that people expect to gather more information in future
about the value of the good, then a person’s WTP today contains an option
value associated with the value of the future information (see section 3.4). If so,
comparing bids for two goods in the full value bidding method entails comparing
an individual’s expected value for the two goods and the option values for
both goods. There is a confound between the compensating variations for the
goods and the option values. If one is only interested in comparing the value
participants place on an attribute (i.e., the compensating variation), and not
the option values associated with future information, the endowment method
presents a straightforward way to eliminate comparisons of option values and
focuses attention on comparisons of valuations between two goods. With the
endowment methods, participants know they will leave the experiment with one
Conducting experimental auctions 67
of two goods, either X or Y, thus the opportunity for future learning about both
goods is eliminated.
In addition to controlling for the problem of field substitutes and option
values, the endowment method has several other advantages. The endowment
method is useful because it can be used with a consumption requirement as
in Shogren et al. (1994). When a consumption requirement is imposed, peo-
ple must consume the good they possess at the end of the auction. Winners
of the auction consume the auctioned good; losers of the auction consume
the endowed good. Several reasons exist as to why one might want to use a
consumption requirement. First, it increases attentiveness and encourages par-
ticipants to put cognitive effort into their bidding decisions. The requirement
encourages subjects not to submit $0 bids for both products as they might in
the full bidding method. Second, the consumption requirement ensures that the
value elicited corresponds to the value of the person who is bidding. Without
a consumption requirement, a subject might bid on a good to give to a peer or
spouse and formulate their bid based on their expectation of someone else’s
value. Further support for the endowment approach is given by Hoffman et al.
(1993), who contended that the most useful and reliable estimates from experi-
mental auctions are differences in willingness to pay (WTP) between two goods.
The endowment approach directly elicits this difference. As illustrated by Lusk
et al. (2001a), the endowment approach can be helpful in attracting participants
in a field setting such as a grocery store. Rather than paying subjects to attend
a laboratory setting, which has been shown to influence valuations (Rutström,
1998), subjects can be given a lower quality good for participation and WTP
for a higher quality good can be elicited. Such an approach removes the effects
of high participation fees.
The endowment approach has some disadvantages relative to the full bid-
ding method. Tversky and Kahneman (1979) suggest that valuations may be
reference-dependent. One manifestation of reference-dependent valuation is
that people might place greater value on a good if they possess it than if they
do not – an effect that is thought to arise from loss aversion, where losses are
valued more highly than gains. If the reference-dependent preferences exist,
then value estimates for a novel product or attribute depend on the subject’s
initial reference point: whether they initially possess the good.
Another problem with the endowment method is that the endowment itself
might send an implicit quality signal to participants. If a person is endowed
with a good and is asked to bid to exchange their endowed good for another,
it might create the impression that the endowed good is in some way inferior
to the auctioned good. The latter concern could be assuaged by initially asking
which good people prefer and endowing them with the lesser preferred good as
in Lusk et al. (2001a).
68 Experimental Auctions
English Sequentially offer Last offered bid Highest bidder pays market 1
ascending bids price
2nd Price Simultaneously Second highest Highest bidder pays market 1
submit sealed bid price
bids
nth Price Simultaneously nth highest bid n − 1 highest bidders pay n−1
submit sealed market price
bids
BDM Simultaneously Randomly drawn People pay market price if individually
submit sealed price bid exceeds randomly determined
bids drawn price
Random nth Simultaneously Randomly drawn n − 1 highest bidders pay n−1
price submit sealed (nth) bid market price
bids
Collective Simultaneously Mean bid Each individual pays market none or all
auction submit sealed price (subject to
bids unanimity rule) if sum of
bids exceeds sum of costs
per se as subjects do not bid against one another in a market environment, but
the structure is nonetheless similar. With the BDM, a person submits a bid and
purchases the good if their bid is greater than a randomly drawn price.
Shogren et al. (2001a) formally introduced the random nth price auction to
combine features of the second price auction, which encourages competition
amongst bidders, and the BDM mechanism, in which there is a relatively high
chance any bidder can win. In a random nth price auction, subjects submit
sealed bids for a good, one bid is randomly drawn from the sample, and all
bidders that bid higher than the randomly drawn bid win the auction and pay
an amount equal to the randomly drawn bid.
Originally introduced by Smith (1980) to value public goods, one could also
use a collective auction. Here, sealed bids are submitted and summed. If the
sum of bids is greater than the sum of the cost of provision (a cost that is
unknown to the subjects when bidding), each person pays a price equal to the
mean bid amount so long as all people agree to proceed with the purchase via
an anonymous vote.
Which mechanisms should be chosen for a given application? There is no
straightforward answer. From a purely theoretical basis, it should not matter; all
auctions should yield bids equal to true values. Auction theory relies on many
assumptions, some more tenuous than others. The choice of auction mechanism
comes down to pragmatic considerations and to properties of auctions that
have been determined through empirical research. We now review some of the
studies investigating the empirical properties of the auction mechanisms to aid
the selection decision.
Evidence exists that subjects over-bid in second price auctions in induced
value experiments (Kagel et al., 1987); but this finding is far from universal
(Shogren et al., 2001a). In an induced value experiment, Noussair, Robin, and
Ruffieux (2004a) found that the second price auction was demand revealing and
generated more accurate bids, as measured by the absolute difference between
bids and values, than the BDM.
In another induced value experiment, Shogren et al. (2001a) used a within-
subject experimental design to compare bidding behavior in the random nth
price and second price auction, where participants were informed of the weakly
dominant strategy in both mechanisms. Overall, their results showed that the
second price auction worked better on-margin (i.e., for people with valuations
near the market price), but the random nth price auction worked better off-
margin (i.e., for people with valuations far from the market price). The two
auctions did not perform differently when data for on-margin and off-margin
bidders were pooled. A regression of induced values on bids generated estimated
lines that were flatter than the perfect-revelation line, with positive intercepts
and slopes below one. In the case of the random nth price auction for on-margin
bidders, this flattening was significant. This behavior could represent a tendency
Conducting experimental auctions 71
for bidders who do not fully understand the random nth price mechanism to
offer bids near the middle of the range. This pattern can arise if the parameter
estimates are biased by the classic attenuation of least squares (Greene 2000,
p. 437) which occurs when an independent variable is measured with error. If
the mean error is zero and no correlation exists between induced value and the
error with which it is measured, the slope coefficient is biased down and the
constant term biased up. In the present context, this would mean the induced
value is transformed inside the subject by some noisy and unobserved process,
for example, a belief that the experience of redeeming tokens will be occasion
for social display or embarrassment.
Whereas Noussair et al. (2004a) compared the second price auction to the
BDM and Shogren et al. (2001a) compared the second price auction to the
random nth, Lusk and Rousu (2006) carried out a comparison of all three
mechanisms (the second price auction, random nth price auction, and BDM
mechanism) in an induced value experiment. Their analysis focused on com-
paring accuracy of the mechanisms by investigating the squared/absolute dif-
ference between bids and induced values. They found that, on average, the
second price and random nth price auctions were significantly more accurate
than the BDM. Consistent with Shogren et al. (2001a), they found a tendency
for the random nth price auction to have a lower mean squared deviation for
off-margin bids than the second price auction, whereas the second price auction
had a lower mean squared deviation at high induced values than the random nth
price auction.
Overall, these studies suggest that choice of auction mechanisms might
depend on whether one is interested in more accurately estimating the upper
or lower end of the demand curve. The second price auction is likely to gen-
erate more accurate bidding for high value bidders; the random nth is likely to
generate more accurate bidding for low value bidders.
In addition to comparisons of mechanisms in induced-value experiments,
several studies have compared homegrown values across auction mechanisms.
Lusk et al. (2004a) found in initial bidding rounds, the second price, random
nth price, English, and BDM all generated similar mean bids for beef steaks.
After several rounds of bidding, however, the second price auction generated
higher bids than the other mechanisms, and the random nth tended to generate
lower bids than the other mechanisms. Rutström ((1998) found that the English
auction and the BDM mechanism generated similar mean bids for chocolates,
but the second price auction generated bids in excess of either of the other
mechanisms. In a field experiment where trading cards were auctioned on the
internet, Lucking-Reiley (1999) found that the second price and English auc-
tions generated similar levels of revenue.
In addition to the empirical findings, other pragmatic factors can come into
play when considering which mechanism to use for a given application. Most
72 Experimental Auctions
people are somewhat familiar with the English auction, and as such, experimen-
tal procedures for this auction should be relatively easy to convey to subjects.
The English auction is also an “open” auction in that every subject knows
the bids of every other subject. This can be an advantage in the sense that
subjects can quickly incorporate market information into their valuations, but
this feature can also be a disadvantage if participants are unduly influenced
by other bidders or if bid affiliation (discussed in a later section) is a major
concern.
The second price auction is relatively easy to explain to subjects and to imple-
ment. Nevertheless, evidence suggests subjects might over-bid in this auction
and there is potential for low-valued subjects to “misbehave” in multiple bid-
ding rounds. The BDM mechanism is the only one that can be used on an
individual basis that does not require a group of subjects (although the mecha-
nism can be used in a group setting by randomly drawing a price for the entire
group). The mechanism has proved useful for eliciting values in field settings
such as grocery stores (Lusk et al., 2001a; Rousu et al., 2005). The strength
of the BDM is also its weakness. Because people participate individually with
the BDM, no active market is present. Shogren (2005) contends that an active
market environment is important in inducing economic rationality at the margin
or at the individual level if arbitrage has enough power. Shogren et al. (2001b)
observed that certain anomalies, for example the WTP/WTA disparity, are still
present with the BDM, but disappear in active market environments such as the
second or random nth price auctions. If a BDM mechanism is implemented,
there are several factors a researcher must consider. For example, what distri-
bution should be used for the random price generator and how should this be
conveyed to participants? It is desirable to use a price range greater than the
range of potential values so that bids are not censored, but care must also be
given that the range is not so large that participants with reasonable valuations
have little chance of winning.
An advantage of the random nth price auction is that it potentially keeps
bidders with relatively low values engaged in the auction. It also provides
a relatively high degree of market feedback if used over multiple bidding
rounds with posted prices. The random nth price auction can be a challenge to
explain to some participants and can be difficult to manage for the experimental
monitor.
One factor to consider when choosing an auction mechanism relates to
the shape of the payoff function and the expected cost from bidding sub-
optimally (Lusk et al., 2006a). Harrison (1989) originally suggested using such
an approach to explain “misbehavior” in first price auctions. Although his sug-
gestion to evaluate behavior in terms of expected payoff was heavily debated and
has some limitations (see Cox et al., 1992), we believe that investigating indi-
viduals’ expected payoffs should be one of several criteria practitioners might
Conducting experimental auctions 73
E πi
(N − 1)! N −k
bi (F(v)) [1 − F(v)] k−2
f (v)
(N − k)!(k − 2)!
= vi − bi vdv
(N − 1)!
−∞ −∞ (F(x)) N −k [1 − F(x)]k−2 f (x)xd x
(N − k)!(k − 2)!
N −1
(N − 1)!
× F(bi )r [1 − F(bi )] N −1−r (5.3)
r =N −k+1
r !(N − 1 − r )!
assuming all bidders except individual i bid truthfully (i.e., b j = v j for all
j = i). While F represents the distribution of random prices in the BDM, which
is an endogenous choice to the researcher, it represents the distribution of bid-
ders’ values in a kth price auction, which is exogenous to the researcher.
74 Experimental Auctions
a relatively high value bidder (vi = 7). If a person with true value of $7 bids
$8 instead of $7, their expected loss in profits is $0.13, $0.07, and $0.05 for
the third price, second price, and BDM mechanisms, respectively. Figures 5.1
and 5.2 both show that in the second and third price auctions, it is more costly
to overbid than to underbid, but that the costs of deviation are symmetric for
the BDM. These findings rest on the assumption of a uniform distribution
and eight bidders, which is likely to vary in each study. The expected costs
of sub-optimal bidding can be calculated for a wide variety of mechanisms,
distributions, participant numbers, and values given the preceding equations.
Whether the equations are actually used to calculate the expected cost of
submitting bids that differ from true values, the general ideas presented are
important to consider. For example, does an auction mechanism and do auction
instructions convey that each person has a chance of winning? If an individual
believes they are not likely to win, they have little incentive to bid truthfully.
The figures also suggest that mechanisms differ by the type of individual that
is given the greatest incentive for truthful bidding. While the BDM imposes
higher costs for non-truthful than the second price auction for low-value partic-
ipants, the opposite is true for higher-value people. Practitioners might consider
which type of person is of most interest. The figures also show that while the
BDM imposes equal costs for under- and over-bidding, the second price auction
76 Experimental Auctions
imposes higher costs for over-bidding than under-bidding. (It is also true that
a kth price auction with k = N − 1, e.g, an seventh price auction with eight
people, would impose higher costs for under-bidding than for over-bidding.)
Practitioners might consider whether, for their particular application, people
are more likely to use heuristics to over- or under-bid and use a mechanism that
counters such behavior.
from an additional unit of a good falls as more units are consumed. If two
substitute goods are simultaneously auctioned, there is a chance a person could
end up purchasing two goods, which would be similar to obtaining two units of
a good. To illustrate, consider the data reported in Corrigan and Rousu (2006a).
They found that when subjects bid for only one jar of salsa, the average bid was
$0.65, but when subjects bid to purchase two jars of salsa, the average bid was
$1.13. Demand for the second jar of salsa was effectively $0.17 (or 26%) lower
than the first unit. Intuitively, we might expect a person bidding in an auction
in which they might win two units of a good to reduce their bid for each good
if they expect they might win both.
In the previous section, we showed the payoff function associated with the
BDM mechanism. Reconsider this auction payoff function for the case where
prices are drawn from a uniform distribution on [0, T]:
b
E[π ] = [v − 0.5b] . (5.6)
T
Now consider an experiment in which a person bids simultaneously for two
substitutable goods in two concurrent BDM mechanisms both using a uniform
random price generator on [0, T]. Let d represent the money-metric discount a
person receives from having two units of a good versus only one that results from
demand reduction or diminishing marginal utility. A person’s payoff function
for bidding on two goods (A and B) simultaneously is:
bA bB bA bB
E[π] = [v A − 0.5b A ] + [v B − 0.5b B ] −d , (5.7)
T T T T
where b A and v A represent the bid and value for good A, respectively, and bB
and vB represent the bid and value for good B, respectively. The last term,
A B
d bT bT , represents the discount from diminishing marginal utility, d, multiplied
by the probability of winning both goods given bids b A and b B . Differentiating
the equation with respect to b A and b B and setting equal to zero, the payoff
is maximized when b A = v A − d(b B /T) and b B = v B − d(b A /T). Instead of
bidding true value, a person offers a bid equal to his value for the good less the
diminishing marginal utility discount factor times the probability of winning the
other auctioned good. After a bit of algebra, we see the optimal bid for good A is:
∗ T (v A T − v B d)
bA = . (5.8)
T 2 − d2
The payoff function is not maximized at b A = v A as in a single good valua-
tion unless A and B are completely unrelated in consumption, that is, d = 0.
The differences in bids are contaminated by demand reduction. Further alge-
∗ ∗
braic manipulation yields the insight that at optimum b A − b B = (v A − v B )/
(1 − d/T ). The difference in bids for goods A and B is not equal to the difference
78 Experimental Auctions
in value between goods A and B. In most instances, T > d, and as such, dif-
ferences in bids between goods will tend to be greater than differences in val-
uations.
This problem is exacerbated if subjects bid on multiple goods in several
bidding rounds and have the possibility of winning goods in each bidding
round. To further illustrate, consider an application in which a subject bids on
two unrelated goods, but in which they must submit bids on each good in each
bidding round. Consider a participant who wins a unit of good A in round 1, but
does not win a unit of good B. Now, demand for good A will decline by d in the
second round, but demand for good B will be unchanged. In round 2, the value
of good A versus good B has changed due to a movement down the good A
demand curve; not necessarily due to an inherent difference in the value of the
goods. If d is sufficiently large, a person that prefers good A in round 1 could
become a good B preferring individual in round 2. The inferential problem of
demand reduction grows more complex if one allows goods A and B to be
related in consumption within any round. The key issue here is that research
results might be an artifact of poor experimental design instead of a reflection
of true preferences that would be displayed in the marketplace.
This raises the question of how to handle demand reduction and diminishing
marginal utility. First, demand reduction is not a problem per se; it is an essential
feature of consumer preference. The problem lies in drawing correct inferences
when demand reduction exists. If one wants to determine how demand falls
as multiple units are purchased, there are multiple-unit variants of the Vickrey
auction that are incentive compatible (see Vickrey, 1961; List and Lucking-
Reiley, 2000). For example, suppose 10 units of a good were offered for sale.
A multiple-unit Vickrey auction would entail each individual submitting up to
10 bids for each unit, with the top 10 bids winning the auction and all 10 units
being sold at the 11th highest bid amount. See Brookshire et al. (1987) for an
application of such an auction in a food marketing context.
The more traditional case is in estimating a person’s value for one unit of
a good (or a unit each of a few differentiated goods). One solution is to con-
duct an auction for a single unit of a single good in a single bidding round.
This solution is overly restrictive because it eliminates within-subject valuation
comparisons and does not allow the researcher to easily investigate the value
of multiple product attributes. A straightforward way to mitigate the demand
reduction problem is to use random drawings to determine a binding bidding
round and/or a binding good. For example, Hayes et al. (1995) had subjects
participate in twenty bidding rounds, in which only one of these rounds was
drawn as binding; Lusk et al. (2004a) auctioned five different types of beef
steaks and randomly drew only one of the steak types as binding after all bids
were submitted. Assuming a subject’s expected utility is linear in probabili-
ties, randomly drawing a binding auction should produce the same results as
Conducting experimental auctions 79
bids increased from $25.2 to $34.4 to $48.6. Overall, the evidence suggested
bidders considered the outside option when formulating their bid strategies, and
changed their bids accordingly. Results also indicate that although bid reduction
existed, bidders who reduced bids did not necessarily bid exactly the price of
the outside option. Statistical tests found both real and hypothetical bidding
schedules were significantly above the choked demand curve. Finally, while
bidding was subject to hypothetical bias, bid shaving due to the outside option
was equally likely to occur in real and hypothetical auctions.
These findings suggest that practitioners need to think about: (a) controlling
for the prices and availability of field substitutes; or (b) making this feature
an integral part of the experimental design. A simple method of control is to
provide subjects with a common expectation about the price of outside options.
For example, Lusk et al. (2004a) explicitly informed subjects of the price of
one of their steaks in the local grocery store. Harrison, Harstad, and Rutström
(2004) discuss methods for dealing with bids that may be truncated at the stated
field price if such a procedure is implemented. Another method of control is to
ask people what they believe to be the price and availability of outside options
and to use this information in subsequent regression analysis to help explain
bidding behavior. This approach was partially implemented by Corrigan (2005).
The task of directly incorporating field substitutes into the research design is
more complicated, but one option might entail offering a menu of potential
substitutes at a posted price while auctioning another good of interest. Varying
the price of these substitutes in repeated auctions would provide insight into
the degree of substitutability between goods.
Figure 5.3 Second price auction bids for five beef steaks across five bidding
rounds
Based on this theoretical literature, Harrison et al. (2004) argue that values
become affiliated because posted prices serve to signal relative quality or the
price/availability of outside options. People take the posted prices as signals of
what the good should be worth to themselves, causing values to become inter-
related or affiliated. That auction bids tend to increase each bidding round in
homegrown value auctions provides prima facie evidence of bidder affiliation.
But such a phenomenon could also be related to learning about the mechanism.
People can be unfamiliar with the types of auctions we are considering. In auc-
tions and in everyday bargaining experiences, people have probably developed
heuristics such as “buy low” that they might follow even if told that optimal
strategy is to bid true value. Repeated rounds with price feedback can force
people to abandon such heuristics, but it might also cause affiliation as prices
send signals of unknown characteristics of the good.
The difficulty in determining the veracity of the criticism of price feedback
is that true values are not known in homegrown value experiments and it is
impossible to know whether people are bidding true value or whether values
are affiliated. It is also a challenge to separate learning effects from affiliation
effects. A few studies have tackled the issue of affiliation directly in homegrown
value auctions. List and Shogren (1999) found that repeated auction rounds
with price feedback caused a slight increase in bids for unfamiliar goods. They
used panel data from published and unpublished experimental auctions that
used repeated trials of the second price auction for familiar and unfamiliar
products such as food products that varied in safety (e.g., sandwiches with
a risk of food-borne diseases or produced with growth hormones). All the
experiments followed a common experimental design. The instructions said
why it was in a bidder’s best interest to tell the truth when bidding in the second
price auction. Each experiment used a repeated round design, in which only
one round was chosen at random as the binding auction. The second-highest
price was publicly posted after each round. The posted price provided subjects
with market price information generated by the group interaction that would
otherwise be unavailable in one-shot experiments.
Table 5.2 summarizes the six categories of data collected: categories 1 and 2
represent naı̈ve willingness to pay (WTP) bids and willingness to accept (WTA)
offers for unfamiliar products without non-price information about the product;
categories 3 and 4 represent informed WTP bids and WTA offers for unfamiliar
goods with non-price information; and categories 5 and 6 represent ex ante
WTP bids and WTA offers for a familiar good (Snickers or Milky Way candy
bars).
List and Shogren (1999) estimated the following model for each of the six
categories of panel data:
where Bit denotes median bid or offer in experiment i and trial/round t; Pi ,t−1
denotes the posted price in experiment i for trial t − 1, αi represents a
fixed/random experiment effect, ϕt is a trial fixed/random effect, and εit repre-
sents the contemporaneous error term. The median bid was used to proxy for
bidding behavior to avoid extreme outliers and overweighting random patterns
of bidding behavior in the last few trials of experiments. Auction or experiment
effects, αi , control for characteristics that vary from auction to auction but are
invariant over trials of the same auction, for example, unique characteristics of
auctioned goods, the lab environment, and nuances of monitors and subjects.
Trial effects, ϕt , capture variables invariant across auctions, and controls for
trends in subject behavior and market experience.
Table 5.3 summarizes the regression results. The results in the top panel of
Table 5.3 suggest that a posted price effect does exist for unfamiliar products,
although the size of the effect is very small. Estimated coefficients (column 1)
indicate posted prices affect the behavior of the median naı̈ve WTP bidders
facing unfamiliar goods. A dime increase in the posted second price increased
the median willingness to pay by about a penny. Parameter estimates (column 3)
for the naı̈ve WTA measures for unfamiliar goods are similar but even less severe
in economic significance; if the posted price falls by one dollar the median offer
falls by about half a cent. Absent information about the product, naı̈ve bidders
seem to rely on the signal sent by the posted second-price from round t when
updating their bids, or offers, in round t + 1.
Is the tendency for prices to lead naı̈ve bidders to increase bids economically
significant given that the unfamiliar products, sandwiches, would retail for about
$4? If the posted price increased by one dollar between trial two and ten, the
implied increase in the median naı̈ve WTP bid is about eight cents or about 2%
of value. Large price spikes were rare in the WTP naı̈ve markets. The average
84 Experimental Auctions
Table 5.3. Two-way fixed effects estimation results for bid equation
Ex Ante Unfamiliar
posted price increased to $1.59 (s.d. = 1.28) in the late trial from $1.21 (s.d. =
0.76) in the early trial, which results in a point estimate increase of $0.03 in the
median bid (= $0.38 × 0.08), which is less than 1% of value. If we build a 95%
confidence interval around the estimated coefficient, β = 0.08, we see that the
lower and upper bounds of the price effect are $0.01 and $0.05, translating into
about 0.02% and 1.3% of value. This price effect seems relatively modest. About
five cents of the change in median bids is caused by posted prices, implying a
small percentage of overall variability is attributable to price changes.
In the WTA naı̈ve regression, the average posted price falls to $9 (s.d. =
7.21) in trial ten from $8 (s.d. = 125) in trial two. With the small regression
coefficient estimate and the relatively large standard error for the WTA bids,
the argument made for naı̈ve WTP holds for WTA as well. The price-induced
decreases are within the margin of error surrounding the average bid, suggest-
ing an exaggerated fear of onerous quantitative bias for the goods considered
herein.
Table 5.3 shows that the price effect disappears when the subject is given
non-price information or is more familiar with the product. Bidders seem to use
price and non-price information as substitutes when formulating their measure
Conducting experimental auctions 85
of value. Results indicate that posted prices affect neither the median informed-
WTP bid nor the median informed-WTA offer at any conventional level of
significance (columns 2 and 4). This suggests that informed bidders treat the
written word as a substitute for market signals. Bids and offers for the familiar
candy bars reinforce this observation; posted prices do not affect WTP bids or
WTA offers (columns 5 and 6) at conventional significance levels. In general, the
more a bidder knows about the good before entering the lab, the less the posted
second price affects his or her behavior. List and Shogren (1999) concluded
that the ultimate effect of repeated rounds and price-posting was to improve
learning about the mechanism.
Alfnes and Rickertsen (2003) also found that posted prices caused small
increases bids. They also showed that the trend in bid increases was similar
across all goods investigated in their study leading them to conclude that learn-
ing about the mechanisms is primarily responsible for the increase in bids over
bidding rounds.
Shogren et al. (2001b) also provide support for multiple-round bidding with
price feedback showing that the procedure yields valuations more consistent
with neoclassical economic theory. Neoclassical theory suggests that the WTP
and WTA measures of value should be similar under certain assumptions; other
studies have shown significant divergence between the two measures. Consis-
tent with these studies, Shogren et al. (2001b) showed that WTA was 1.5 to
2 times WTP in the first rounds of bidding; but, the two measures converged
over repeated rounds for auction mechanisms that used active markets with
price feedback. We describe this study in more detail in Chapter 9.
Harrison, Harstad, and Rutström (2004) re-analyzed the data in Hoffman
et al. (1993) and showed that controlling for price feedback in regression anal-
ysis reversed one of the main conclusions of the original study regarding the
relative value of vacuum packaging. They found that bids for beef steaks were
affected by prices in previous rounds and they argued that such an effect was due
to affiliation. While also consistent with standard notions of Bayesian updating
on beliefs, this finding led them to argue against using repeated round auc-
tions. But before one accepts this conclusion, recognize that Hoffman et al.’s
(1993) experimental design made every round binding. The finding that prices
in previous rounds affected subsequent behavior is not surprising as it is due
to movement down the demand curve as people purchase additional steaks and
demand curve shifts as people learn the price of a substitute good. Because
of this feature of Hoffman et al.’s experimental design, repeated rounds are
likely to have generated a much larger effect on behavior. Homegrown value
studies after Hoffman et al. (1993) have selected only one round to be binding
circumventing the confounding influences on bids.
One can also think about this issue from a slightly different angle by com-
paring bidding behavior across mechanisms that use different levels of price
86 Experimental Auctions
feedback. For example, Lusk et al. (2004a) and Rutström (1998) both found
that the BDM mechanism and English auctions generated statistically equiva-
lent results. This finding is fascinating because the English auction permits the
greatest possible amount of market feedback as every person can see exactly
what every one else is doing (and has the greatest potential for bidder affilia-
tion), whereas the BDM permits no feedback about other individuals’ values
(and has no bidder affiliation). Two mechanisms which span the spectrum in
terms of potential of bidder affiliation seem to generate very similar results in
single-round auctions. These results seem to suggest that changes in bids across
bidding rounds are unlikely to be due to affiliation.
We now consider some new findings on this question. The difficulty with
many previous studies investigating bidding behavior over repeated rounds is
that affiliation is defined as a positive relationship between posted prices in a
previous period and bids in the current period. Such an effect could also result
from learning about the mechanism. For example, suppose a person, based on
previous buying behavior in other contexts, takes a strategy of under-stating their
true valuation in a misunderstood attempt to garner additional surplus. Posting
a high price will cause such a person to re-evaluate this strategy and force them
to increase their bid and the higher the price the more pronounced the departure
from the heuristic. A positive relationship between posted prices and bids need
not imply affiliation. We take a different approach to investigate whether the
practice of posting prices in repeated round auctions causes affiliation. Recalling
a primary argument against price feedback is that people use posted prices to
partially infer the quality of the good, we investigate whether relative valuations
for qualities of a good are affected by posted prices.
Looking at Figure 5.3 yields some insight into this issue. While bids for
each steak type tend to increase across rounds, the differences between steak
bids remain reasonably constant as the lines are roughly parallel. Relative pref-
erences for quality do not seem to be changing with the price information,
which implies people are not using posted prices to signal unknown quality or
availability. The data in Table 5.4 provide more formal support for this obser-
vation. Parametric and non-parametric tests reject the hypothesis that mean
bids for each steak are unchanged over the five bidding rounds. When inves-
tigating the differences between each steak’s bid and the average bid for all
steaks, however, we cannot reject the hypothesis of uniform bids across rounds.
While the mean bid for the generic steak increased about $1.20 from round
one to round five (an increase which is statistically significant), people bid
$0.79 less for the generic steak than the mean for all steaks in round one and
this statistic only increased one cent to $0.80 by round five (not statistically
different).
To delve even further into the issue, Table 5.5 reports regression results
similar to those in List and Shogren (1999), Alfnes and Rickertson (2003), and
Harrison et al. (2004), in which bids are regressed against dummy variables for
Conducting experimental auctions 87
Table 5.4. Second price auction bids for beef steak across five bidding rounds
Absolute bids
($/12 oz steak)
Generic $2.46 $3.06 $3.29 $3.50 $3.66 0.07 0.05
Guaranteed tender $3.29 $3.88 $4.14 $4.22 $4.26 0.26 0.05
Natural $2.91 $3.57 $3.71 $3.74 $4.03 0.23 0.15
Choice $3.60 $4.37 $4.68 $5.01 $5.15 0.01 0.00
Certified Angus Beef $3.97 $4.71 $4.97 $5.18 $5.17 0.09 0.04
Overall mean $3.25 $3.92 $4.16 $4.33 $4.45 0.05 0.03
Differences from
overall mean
Generic −$0.79 −$0.86 −$0.66 −$0.68 −$0.80 0.99 0.96
Guaranteed tender $0.04 −$0.04 $0.07 −$0.07 −$0.19 0.76 0.99
Natural −$0.34 −$0.35 −$0.42 −$0.31 −$0.43 0.89 0.86
Choice $0.36 $0.45 $0.85 $0.82 $0.70 0.14 0.37
Certified Angus Beef $0.73 $0.79 $1.02 $0.84 $0.72 0.95 0.81
a p-values associated with the hypothesis that the mean bids (central tendencies in the case of the
Rank-Sum test) for each steak are equal across all five bidding rounds.
each steak type, dummy variables for each round, and the lagged price posted
in the previous round. When aggregate models are estimated, as in previous
studies, we find posted prices affect both absolute and relative bids.2 Results
indicate a $1 price increase in a previous round increased bids in the subsequent
round by about $0.30; similarly a $1 increase in the price of a steak relative to
the mean price for all steaks in a previous round increased bids for the steak by
about $0.30 over the mean for all steaks in the subsequent round. At first blush,
these results are largely consistent with previous findings. The problem with the
aggregate models is that they are mis-specified. F-tests reveal strong interactions
between all the variables in Table 5.5 and individual dummy variables, that is,
we reject the hypothesis that all people have the same coefficients for each
variable in Table 5.5 at the 0.05 level of significance.
Because the thirty-five people in our sample submitted twenty-five bids (each
person bid for five steaks in five bidding rounds), we are able to estimate
individual-level models. Table 5.5 reports the average coefficients from these
thirty-five models. Overall, results are similar to the aggregate models except
the lagged price effects are not statistically significant at any standard level of
2 The aggregate models are ordinary least squares regressions including fixed effects for each
individual; less than 4% of the bids were zero and as such using a censored model instead of the
OLS has virtually no effect on results.
88 Experimental Auctions
Table 5.5. Aggregate and individual models of the effect of posted prices on
bidding behavior
Average Average
parameters from parameters from
Pooled model with individual-level Pooled model with individual-level
Variable fixed effectsa modelsb fixed effectsc modelsd
a Number of observations = 700; fixed effects for each individual were estimated but are not reported
in the table.
b A regression with 20 observations was estimated for each of the 35 people in the sample.
c Number of observations = 560; fixed effects for each individual were estimated but are not reported
in the table.
d A regression with 16 observations was estimated for each of the 35 people in the sample.
e One asterisk (*) indicates statistical significance at the 0.05 level or lower.
f Numbers in parentheses are standard errors.
g Estimated effects relative to the Certified Angus Beef Steak.
h Estimated effects relative to round 5.
i In the models with bid differences as the dependent variable, the lagged price difference is included
as the regressor.
Conducting experimental auctions 89
Figure 5.4 Bid functions and the determination of winner, loser, and market
price
(Bid-High(You)). These two bids were combined to create a bidding line (see
Figure 5.4).
Step 3: Auction winner. The winner of the auction is determined by the
intersection of the two bidding lines relative to the bids are equal line in
Figure 5.4. Consider two bidders: A and B. If Figure 5.4 is drawn from A’s
perspective, A will win because the two bidding lines intersect above the bids
are equal line.
Step 4: Auction price. The price is also determined by the bids are equal
line. For an auction winner, the price is set by where the other player’s bidding
line intersects the bids are equal line.
Step 5: Payoffs. For an auction winner, the payoff for that round is simply:
Resale Value(You) – Auction price. For auction losers, the payoff was $0.
Margolis and Shogren (2004) found, based on regression estimates of private
values regressed on bids, that behavior in their experiment was not perfectly
described by the optimal bidding strategy. The bidding behavior, however, was
systematic. In almost every treatment, the coefficient on private value was below
one. Overall, this suggests bidders did not respond as strongly as they should
to their private signals; rather, they offer bids that were about what the average
Conducting experimental auctions 91
subject offered. While this behavior might indicate that bidders lack confi-
dence in their understanding of the auction and preferred to avoid behavior
that appeared extreme, it is consistent with findings in other auctions of flatter
bidding than the perfect-revelation line. Overall, the results from this initial
exploration suggest this relatively complicated auction designed with inexperi-
enced bidders and no feedback can work to alleviate concerns with affiliation.
Future work in which bidders gain experience through many trials and are pro-
vided more information on payoffs through computer implementation, is likely
to generate greater efficiency levels for the experimental auction.
If the argument is that values are somehow not affiliated when people walk
into an experiment, and that posting prices causes affiliation, conducting only
one bidding round, as suggested by Harrison et al. (2004) has some merit.
But even if people do not react to posted prices, evidence suggests bidding
behavior changes across rounds; an effect most likely due to learning. If only one
bidding round is conducted, training people with the mechanism is extremely
important. One approach is to follow the steps outlined in Plott and Zeiler
(2005) to avoid subject misperceptions. These steps include a significant about
of discussion about the mechanism and numerous practice bidding rounds with
lotteries.
Even if bidders’ values become affiliated over multiple bidding rounds, there
are several interesting areas of investigation that need more attention. For exam-
ple, in everyday shopping experiences, people respond to price information and
other cues from friends, family, and advertising. Affiliation may be a common
feature of individuals’ valuations even when the context has nothing to do with
experimental auctions. It is also worth thinking about why and how values
become affiliated. For example, Lusk et al. (2006) conducted fifth price auc-
tions in several US and European locations, in which they elicited the minimum
amount of compensation subjects demanded to consume a genetically modi-
fied food. These values were elicited over ten bidding rounds, with one of the
rounds randomly selected as binding. As is common in WTA auctions, they
found median bids were relatively high in the first bidding round and trended
downward as the experiment progressed. The trend in bids, however, differed
significantly from location to location and from individual to individual. The
aggregate behavior implies something about the relative importance of mar-
ket prices and individuals’ concern for biotechnology. Subjects, on average,
demanded less compensation to consume a genetically modified food when
they were informed of the market price potentially indicating that a portion
of acceptance/rejection of biotechnology is a function of their perception of
market acceptance.
That some subjects responded differently than others illustrates that affiliation
has several potential mediating factors. Responses to posted prices can have
meaningful interpretation beyond theoretical concerns with affiliation. The key
is to set up the experiment so effects of interest can be properly controlled
92 Experimental Auctions
to ensure that there are no confounding factors. One approach in this regard
is to use surrogate or confederate bidders to control the market price while
investigating changes in bidding behavior (see Corrigan and Rousu (2006b)).
Another approach is to employ a “menu” of nth price auctions – second price,
third price, fourth price, etc., and investigate how bidding behavior responds to
changes in the magnitude of the price induced by a change in the mechanism.
The jury is still out regarding the appropriateness of using repeated rounds
with price feedback. When price feedback is used, care should be taken through
econometric analysis or experimental design to ensure that price feedback, in
and of itself, is not confounding results. More work is needed to uncover the
conditions under which valuations might become affiliated and the process by
which people learn in experimental auctions given alternative market circum-
stances and information.
standard procedures in induced value experiments, except they use three sets
of positive and negative induced values as the treatments and subjects were
assigned a new induced value in each round.
Their results suggest neither the second price nor the random nth price auc-
tion performed without fault. Bidding in the second price auction was precise
but biased; the highest positive-value bidders tended to overstate their values,
whereas the lowest negative-value bidders tended to understate their values. In
contrast, bidding in the random nth price auction was demand revealing across
all induced values, but it was imprecise; the variance was relatively large and
some bidders submitted very negative bids, seemingly strategically. The average
negative-value bidder tended to understate values in the second price auction,
irrespective of positive or negative induced value, and to overstate values in the
random nth price auction.
What if these results transfer to non-induced lab auctions? For Dickinson
and Bailey’s (2002) second price auction on meat characteristics, the findings
suggest their results are an upper bound on consumers’ actual valuation of the
auction items (as they noted). Though negative bids were relatively rare, actual
consumer negative reaction to some types of information they examined would
be understated, based on the findings of Parkhurst et al. (2004). In contrast,
had Dickinson and Bailey (2002) used a random nth price auction, their results
might have been more demand revealing but more noisy, making it difficult to
attach any statistical significance to important changes in bids across individual
auctioned items – a key objective of their study. The choice of auction can
attenuate or exaggerate subjects’ negative reactions to a new good.
An outside observer using such lab data should note the risk of underesti-
mating the negative reaction to new goods in the marketplace (e.g., consumer
picketing or boycotts). That Parkhurst et al. (2004) found insincere bidding by
people with large positive values and a tendency for people with large negative
values to underbid relative to their values suggest that a homegrown value study
with a second price auction might cause one to overstate the extent of public
support and acceptance for a product and understate the extent of opposition.
However, people in the lower tail of the value distribution are most likely to
boycott and protest undesirable products. Kalaitzandonakes and Bijman (2003)
argue, with respect to genetically modified food, it is the extreme consumers
not the average consumer that drives the market acceptance of controversial
products.
The potential for negative values is important for policy analysis. While it
could be argued that people will not have a negative value for a good that they
could freely dispose, for example, a genetically modified food can be thrown
away by someone, policy issues arise in which interest lies in determining the
welfare effects of moving from one state of the world to another. Consider the
welfare effects of a policy requiring all food to be irradiated to reduce the risks
94 Experimental Auctions
5.8 Conclusions
This chapter considered six design issues in conducting an experimental
auction: training and practice; endowment of a product versus full bidding;
choosing an auction mechanism; multiple good valuation, demand reduction
and field substitutes; learning and affiliation in repeated bidding rounds; and
negative values. Our goal in this chapter was to raise critical issues so the reader
is aware that his or her choices of experimental auction design can and do affect
the values elicited. Even if preferences for new goods are robust to the path taken
to elicit them, experience shows that values can be context-specific. The key is to
understand whether these revealed values respond to the context in predictable
ways as predicted by economic theory (e.g., lower values given lower priced
outside options) or whether these values react to uncontrolled contextual cues
not addressed or appreciated by the experimenter (e.g., negative values). Given
the nascent nature of the field of experimental auctions, the list of design issues
considered in this chapter is not comprehensive. Future research will identify
additional strengthens and weaknesses of these six features and most likely will
raise concerns about other design issues. Being aware of the potential pros and
cons of the topics discussed herein should help the reader look for additional
challenges and opportunities when conducting experimental auctions.
6 Data analysis
6.1 Introduction
We now examine several methods commonly used to examine bidding behavior
in experimental auctions. We focus attention on econometric models and other
techniques that have direct applicability for auction data and are likely to receive
less coverage in introductory econometrics and statistical texts. The chapter
does not cover elementary statistical analysis such as ANOVA and ordinary
least squares regression since numerous sources already exist for interested
readers (e.g., see Greene, 2000).
95
96 Experimental Auctions
such censored observations, all that is known is that the true value difference
falls in the range [WTP2 − WTP1 , ∞). Similarly, if a person’s bid was zero
in treatment/round 1 but positive thereafter, then the calculated bid difference
may be less negative than the true difference. For such censored observations,
the true bid difference falls in the range of (−∞, WTP2 − WTP1 ]. The point
is that the observed value y is not a direct corollary to true underlying value or
value difference, y∗ .
Let y∗ = Xβ + ε, where X represents a matrix of explanatory variables
hypothesized to influence bids such as attitudes, risk preferences, demograph-
ics, or treatment identifiers, β is a conformable vector of coefficients, and ε
is an error term assumed normally distributed with mean zero and variance
σ 2 . One of four forms of censoring can occur, each of which is discussed in
turn.
Left censored observations (Censoring from below). An observation is left
censored if y = tL when y∗ ≤ tL , where tL is the point of censorship. For
auction data tL = 0 when people are not permitted to bid negative values. In
general, the contribution of left censored observations to the likelihood function
is ((tL − Xβ)/σ ), where is the cumulative standard normal distribution
function. If tL = 0, the contribution of left censored observations to the overall
likelihood function simplifies to 1 − (Xβ).
Right censored observations (Censoring from above). An observation is right
censored if y = tR when y∗ ≥ tR , where tR is the point of censorship. If a familiar,
pre-existing good is auctioned in a lab setting, tR might represent the price of the
good outside the lab as people would not rationally bid above the price for which
they can easily buy the good when the session is over. An auction bid would
represent true values (y∗ ) at values below tR , but will only represent a lower
bound when bids equal tR . The contribution of right censored observations to
the likelihood function is 1 − ((tR − Xβ)/σ ).
Interval censored observations. An observation is interval censored if all that
is known is tI1 < y∗ < tI2 . This situation can arise, for example, if auction bids
are elicited in intervals. Sometimes people are asked to indicate their bid by
checking one of the following intervals, $0 to $0.99, $1.00 to $1.99, $2.00 to
$2.99, etc. The contribution of interval censored observations to the likelihood
function is:
Again, the estimated coefficients resulting from the maximization of (6.2) rep-
resent the marginal effects of X on y∗ , but do not represent the marginal effect
of X on observed data, y. When data is always censored, one might be more
interested in obtaining the mean of y and marginal effects of X on y. As shown
by Greene (2000), the censored mean and censored marginal effects are given
by:
Xi β φ((X i β)/σ )
E[yi |X i ] = Xi β + σ (6.3)
σ ((X i β)/σ )
98 Experimental Auctions
and
∂ E[yi |X i ] Xi β
= β (6.4)
∂ Xi σ
McDonald and Mofitt (1980) provide useful results showing how the marginal
effects can be decomposed into two components: the effect of X on y∗ given
that y > 0 and the effect of X on the probability that y > 0. Cragg (1971)
provides another useful insight by suggesting it may be overly restrictive to
assume equality of coefficients in the uncensored and censored portions of the
likelihood function. For example, a particular independent variable, such as
age, might be expected to have a positive effect on the probability a person
bids zero, but a negative effect on observed bids. Even in situations in which
the anticipated signs of the two effects are equivalent, the relative magnitudes
might differ. In its basic form, Cragg’s double hurdle model can be estimated
in a two step process by first estimating a binomial probit model to investigate
the determinants of X on the probability that y > 0 followed secondly by a
truncated regression model where X is regressed on positive values of y = y∗ .
These two models can be estimated separately or jointly. The joint likelihood
function for the double hurdle model comprised of the probit and truncated
regression is:
N
LF = (−X i β1 )(1−ti )
i=1
ti
1 yi − X i β2 X i β2
× (X i β1 ) φ , (6.5)
σ σ σ
where ti = 1 when y > 0 and ti = 0 when y = 0. This specification has two
separate coefficients β 1 and β 2 corresponding to the first and second hurdles.
When the same set of independent variables are used in the first and second
hurdles as in equation (6.5), Lin and Schmidt (1984) show the tobit model is a
restricted version of the double hurdle model, in which β1 = β2 /σ . The appro-
priateness of the tobit versus the double hurdle can be tested with a likelihood
ratio test. The likelihood ratio statistic is calculated as:
where LF represents the maximized log likelihood function values for the model
type indicated in the subscript, each of which is estimated independently. The
null hypothesis is that the tobit model is the appropriate specification, i.e.,
β1 = β2 /σ . If the calculated likelihood ratio statistic exceeds the critical chi-
squared value with number of degrees of freedom equal to the number of vari-
ables in X, the tobit is rejected in favor of the double hurdle model.
Data analysis 99
Hurdle 2:
Ordinary least Hurdle 1: Truncated
Variable Definition squaresa Tobita probitb regressionc
a Dependent variable is the bid for cookie C from the study reported in Lusk and Fox (2003).
b Dependent variable takes the value of 1 if cookie bid was positive; 0 otherwise.
c Dependent variable includes only positive cookie bids.
d Numbers in parentheses are standard errors.
e One (*), two (**), and three (***) asterisks represent 0.10, 0.05, and 0.01 levels of statistical
significance, respectively.
the double hurdle model, which consists of probit and truncated regression
estimates.
The data in Table 6.1 can be used to test the appropriateness of the tobit
specification. Using the likelihood ratio test described above, the calculated
likelihood ratio statistic is −2[−54.601 + 67.819 − 25.083] = 23.73, which
can be compared against the 95% critical chi-square value with seven degrees
of freedom, which is 14.07. These calculations indicate that the tobit model is
rejected in favor of the double hurdle model. As shown in Table 6.1, model
specification is not inconsequential. It is only in the last column that one finds
that the value elicitation setting has a significant effect on bidding behavior.
Conditional on bids being positive, the results indicate bids were higher in the
field setting than in the laboratory setting.
40%
35%
35%
30%
Percentage of Bids
25%
23%
20%
15%
10% 8%
7% 7%
6% 6%
5% 3%
1% 1% 1% 1%
0%
.0 .5 .0 .5 .0 .5 .0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 0.0
$0 $0 $1 $1 $2 $2 $3 $ $ $ $ $ $ $ $ $ $ $ $ $ 1
>$
Auction Bid ($/cookie)
Figure 6.1 Distribution of fifth price auction bids for a non-genetically mod-
ified cookie in France
where θ is the quantile and ρ is a weighted absolute value function (also known
as “check” function) defined as follows:
θ (yi − X i β), if yi − X i β > 0
ρθ (yi − X i β) = (6.8)
(θ − 1)(yi − X i β), if yi − X i β ≤ 0.
102 Experimental Auctions
N
minimize ρθ (yi − max(t L , X i β)). (6.10)
i=1
Regressions
Note: Dependent variable is round five willingness to accept bid to consume a genetically modified
cookie among French consumers; data comes from the study reported by Lusk et al. (2006c).
Number of observations in each regression = 97.
One (*) and two (**) asterisks represent 0.10 and 0.05 levels of statistical significance, respectively.
a Numbers in parentheses standard errors.
for five steaks in five bidding rounds). Table 5.5 shows an important difference
in a key variable of interest (the lagged-price effect) depending on whether a
single constant is estimated for each individual or whether the entire parameter
vector is permitted to be individual specific. An F-test was able to reject the
hypothesis that β i = β for the models reported in Table 5.5.
Bid for
Natural E1
b11
Factor 1
Food Safety b12
Bid for E2
Preferences Traceable
b13
Bid for
E3
Generic
b21
Bid for E4
Factor 2 b22 Guaranteed Tender
Taste
Preferences b23
Bid for E5
b24 USDA Choice
Bid for
E6
CAB
Figure 6.2 Hypothetical path model where six auction bids are represented
by two latent factors
approaches such as confirmatory factor analysis allow the analyst to test sta-
tistically whether a particular theoretical model, such as that in Figure 6.2, is
consistent with the observed data.
The exact methods for carrying out factor analysis, which involves methods
for determining how to select the appropriate number of factors, how to estimate
the path coefficients, how to interpret the path coefficients, and so on is a subject
too vast for this book. There are many good references available and many
common statistical packages such as SAS and SPSS support factor analysis.
For good introductory reference guides on the subject see Hatcher (1994), Kline
(2002), or Kim and Mueller (1978).
that describes the goods to be considered in the simulation. This choice set
can contain the number of goods in the auction, J, a subset of the auctioned
goods, or when a WTP function is estimated, any arbitrarily large set of goods.
Once the initial choice set has been constructed, interest lies in calculating the
market share of the jth good given a set of K goods. In the conjoint literature, a
common approach is to use the “first choice” or “highest utility” rule in which
each person is assumed to choose the product producing the highest utility from
a choice set. If equation (6.16) is less than zero for all K goods, it is assumed
that the person would have not bought any of the goods, that is, no choice would
be made. Under the first choice rule, the market share of a particular good j is
calculated by dividing the number of people predicted to choose the product
(i.e., the number of people having the highest utility for good j) by the total
number of people in the simulation.
A second approach is to use the Bradley-Terry-Luce model or the so-called
share of preference model. With this approach, the market share of good j is
calculated by dividing the utility of good j by the sum of utilities for all K goods
in the simulated market as shown in (6.17):
Uij
Market Shareij = . (6.17)
K
Uik
k=1
The final market share estimate is calculated by averaging each person’s esti-
mated share.
A final model frequently used in market share simulations is the logit model.
Market shares are calculated by dividing the logit value for one product by the
sum for all other products in the simulation. Using the logit formula, the market
share of good j for individual i is given by:
eλUij
Market Shareij = , (6.18)
K
λU
e ik
k=1
good exceeds the bid for each good. This is realistic in that not all consumers
make a purchase every chance they have and that price discounts encourage
additional consumers to make a purchase. If it can be assumed all people make
a purchase, one can also use bids from auctions where people bid to exchange
an endowed good for several substitute goods to calculate market share. A
market share model based on bids from an endow-and-upgrade approach is
essentially a model of utility and price differences. For example, assume people
were endowed with good A and bid to exchange A for goods B and C. Under the
first choice or highest utility rule, a person would be expected to choose good
B if (WTPAB − pB + pA ) was greater than zero and was greater than (WTPAC −
pC + pA ), where WTPAj is the auction bid to exchange good A for good j,
and pj is the price of good j. Good A would be expected to be chosen if both
(WTPAB − pB + pA ) and (WTPAC − pC + pA ) are less than zero.
Once market share has been calculated under a base-line condition, any
number of alternative scenarios can be considered. For example, one might be
interested in determining the market share of a new product and how shares
of existing products will fall once the new product is introduced. As another
example, own- and cross-price elasticities of demand can be calculated by
investigating how market shares change when prices are altered.
6.6 Conclusions
This chapter discussed several data analytic methods useful for investigating
bidding behavior and drawing inferences from auction bids. Some of the most
interesting and useful types of data analysis are the simplest. A great deal can be
learned simply by plotting the distribution of bids and comparing bid distribu-
tions across goods and/or treatments. Similarly, simple comparisons of statistics
such as the mean, median, and standard deviation can reveal important insights.
This chapter considered more advanced methods frequently used in examining
auction data. Since auction bids can be censored from below at zero or from
above at a field price, we covered censored regression models, for example, the
Tobit model. Because auction data are frequently skewed, practitioners may be
interested in the effects of a co-variate, say a treatment dummy variable, on
other portions of the bid distribution than the mean. Quantile regressions per-
mit just such an investigation. We have also considered panel-data regressions
because multiple auction bids are often submitted by the same person. Finally,
we offered some thoughts on other data analytic methods that might have use
in examining auction data, but have yet to be applied to such a context: factor
analysis, cluster analysis, and market share simulations.
7 Valuation case studies*
7.1 Introduction
We now present nine case studies to illustrate experimental auctions at work
in practical applications. This chapter presents case studies showing how we
have used experimental auction methods to estimate the value of new or non-
market goods and how those values are used in welfare analysis and to estimate
the success of new products. These case studies are grouped into three broad
categories: informing policy, marketing, and valuing controversial goods. In
the policy case studies, we consider experimental auctions designed to address
a grading system for beef tenderness, behavior toward food safety, and the
acceptable tolerance levels for genetically modified food. For the marketing case
studies, we present auctions designed to evaluate the market share for branded
beef products, value characteristics of pork, and value financial records. The last
set of case studies focus on controversial goods, where we estimated demand
for genetically modified food, the impact of information on preferences for
irradiated foods, and the demand for products of ambiguous quality like fresh
meat produced with growth hormones.
∗ This chapter draws on research conducted jointly with our colleagues over the last decade includ-
ing Dermot Hayes, John Fox, Matt Rousu, Wally Huffman, Bryan Melton, Abe Tegene, S. Y.
Shin, Jim Kliebenstein, Ted Schroeder, Lisa House, Carlotta Valli, Sara Jaeger, Melissa Moore,
Burt Morrow, and Bruce Traill.
113
114 Experimental Auctions
where αij is the fixed amount individual i is willing to pay for brand j
independent of the steak’s tenderness level. This value can be derived from other
perceived palatability attributes such as juiciness or from utilitarian value, say
from stew meat. Let VTi be the monetary value a person places on tenderness,
φij is the probability that steak option j is tender, and pj is the price of steak
brand j. A consumer is willing to pay αij + VT i for a guaranteed tender steak,
and only αij for a guaranteed tough steak. At present there is currently no grad-
ing system such that steaks can be labeled as to their tenderness. Consumers
make their own subjective assessment about φij , which we label as φ̃ ij . If there
are J steak brands, consumers choose the brand that generates the highest util-
ity. If no steak generates positive utility, the consumer chooses not to buy and
receives the utility of zero. Consumer surplus is given by:
More complicated grading systems can be introduced. One can use a grading
system that categorizes steaks into either high (H) or low (L) tenderness grades.
For each steak brand j there are two products, H and L, with the only difference
being the probability the steak is tender. Let φ Hj and φ Lj denote the objective
probabilities of tenderness for the high and low categories for brand j. Under
such a system (denoted 2G for two grades), consumer surplus is given by:
CSi2G = max αi1 + φ1H VT i − p1H , αi1 + φ1L VT i − p1L , . . . , αiJ
+ φ JH VT i − p JH , αiJ + φ JL VT i − p JL , 0 . (7.5)
and the change in consumer surplus from introducing a grading system with
two grades is:
Once individual-specific values for αij and VTi are calculated, the utility deri-
ved from a steak of any tenderness level is determined through equation (7.1).
We pooled bids from all four auction mechanisms. For the second and nth
price auctions, which were repeated round auctions, we use the bids from the
initial bidding round. We do this for two reasons. First, this case study is meant
to illustrate how auction bids can be used and pooling the data helps ease
the exposition. Second, all mechanisms are theoretically demand revealing
and each has its own advantages and disadvantages. A literature exists that
shows that composite forecasts from various models or people generally out-
perform a single forecast (e.g., Clemen, 1989). We believe it is reasonable
to expect that estimates of WTP (e.g., forecast of true WTP in the popula-
tion), from several mechanisms might outperform estimates from any single
mechanism.
7.2.4 Results
In total, 119 subjects took part in the auctions; three people did not provide
usable observations on the likelihood the generic steak was tender and were
dropped. Table 7.1 contains summary statistics on the bid distributions for each
of the five steaks and the subjective probabilities of tenderness for the generic,
Choice, and CAB steaks. On average, subjects were willing to pay $2.14 to
obtain a 12 oz generic steak, which was $0.81, $1.65, and $2.21 less than the
average bids for the guaranteed tender, Choice, and CAB steaks. The subjective
probabilities of tenderness varied, but averaged 46%, 79%, and 82%, for the
generic, Choice, and CAB steaks.
The estimated value of tenderness, VTi , was calculated for each person and
summary statistics are also reported in Table 7.1. This figure was calculated
by assuming the generic steak and the guaranteed tender steak were identical
except for the guarantee of tenderness in the latter; this assumption along with
equation (7.11) gives the value for VTi . The value of tenderness ranges from
a low of $0.00 to a high of $8.00, with a mean of $1.54. The last three rows
of Table 7.1 show the value of the generic, Choice, and CAB, net the value of
tenderness. On average, the value that consumers derived from a generic steak
that is not a result of tenderness is $1.41; whereas the value that consumers
derived from the CAB steak that is not a result of tenderness is $3.06. This
implies that CAB steaks have other quality attributes that consumers desire
above and beyond tenderness.
We investigated the consumer welfare effects of three different tenderness
grading systems. As a baseline, we calculate consumer surplus under the present
situation, in which only subjective probabilities are used to evaluate steak ten-
derness. In the second scenario, we calculate consumer surplus under the sit-
uation in which steaks are labeled with the average probability of tenderness,
which we set at the average levels shown in Table 7.1. The difference between
the baseline scenario and the second scenario is that in the former each person
uses their own subjective probability whereas in the latter scenario all consumers
Valuation case studies 119
Table 7.1. Summary statistics of auction bids and the value of tenderness
(n = 116)
25th 75th
Variable Minimum Percentile Mean Median Percentile Maximum
WTP for generic steak $0.00 $1.00 $2.14 $2.00 $3.21 $6.75
($/12 oz steak)
WTP for guaranteed tender $0.00 $1.75 $2.95 $3.00 $4.00 $12.00
steak ($/12 oz steak)
WTP for natural steak $0.00 $1.00 $2.59 $2.50 $4.00 $10.00
($/12 oz steak)
WTP for choice steak $0.00 $2.06 $3.79 $3.75 $5.00 $13.00
($/12 oz steak)
WTP for CAB steak $0.00 $2.60 $4.35 $4.00 $5.50 $15.00
($/12 oz steak)
Probability generic steak is 5% 30% 46% 50% 50% 95%
tender (φ̃ generic )
Probability choice steak is 25% 70% 79% 80% 90% 100%
tender (φ̃ Choice )
Probability CAB steak is 30% 75% 82% 85% 90% 100%
tender (φ̃ generic )
Value of tenderness (VT) ($/12 $0.00 $0.00 $1.54 $1.22 $2.04 $8.00
oz steak)
αgeneric = αguaranteedtender −$4.00 $0.00 $1.41 $1.50 $2.76 $5.52
($/12 oz steak)
αChoice ($/12 oz steak) −$3.20 $1.03 $2.55 $2.59 $4.06 $7.09
αCAB ($/12 oz steak) −$3.20 $1.25 $3.06 $2.91 $4.52 $13.04
use a common belief about the tenderness of each steak. In the third scenario,
consumer surplus is calculated when a tenderness grading system is imple-
mented with two grades: high (with a probability of tenderness equal to 85%)
and low (with a probability of tenderness equal to 15%). A fourth scenario
investigates a three-grade system that uses the previous high and low grades,
but with an additional medium tenderness grade with a probability set to 50%.
Table 7.2 reports the price assumptions used to calculate the welfare effects.
Prices were set to correspond to changes in tenderness. For example, in the base-
line scenarios, the price of generic was set at $2.50 and the price of Choice was
set at $3.50. When assigning a price for the “high” tenderness grade, the generic
price was increased $0.50 and the price of Choice was increased $0.25; this was
done because there is a much larger difference between the generic probability
of tenderness between the baseline and “high” scenarios (43% versus 85%)
than for Choice (only 75% versus 85%). In other applications, practitioners
might consider constructing supply functions rather than assuming exogenous
prices. Exogenous prices are used in this case study for ease of exposition and
to illustrate the points at hand.
120 Experimental Auctions
25th 75th
Variable Minimum Percentile Mean Percentile Maximum
Table 7.3 shows the estimates of changes in consumer surplus. Recall the
welfare calculation permits people to opt out of purchasing a steak at all if
none provide positive surplus. This means the units of the welfare measures
are in dollars per choice occasion, that is, each time a person considers pur-
chasing a steak whether they actually consummate the purchase. The number
of shopping trips a person makes in a year might be considered the number of
purchase occasions. Results indicate that moving from no tenderness grading
to the single average grade actually results in a decline in consumer surplus
(about $0.01/choice occasion). This occurs because as shown in Table 7.1, the
median probabilities of tenderness exceed the means for all three steaks; imple-
menting the single average label implies a decrease in the perceived level of
tenderness for more than half the consumers. The next two rows in Table 7.3
Valuation case studies 121
7.2.5 Conclusions
With a grading system in place, consumers who are unconcerned about ten-
derness can find lower-priced steaks that fit their needs; whereas “tenderness-
lovers” can find what they desire. The consumer welfare improvement moving
from two to three grades is minimal; however, the aggregate effects of mov-
ing from no-grading to a two-grade system are substantial. Assuming each US
household makes one trip to the grocery store each week, the aggregate increase
in consumer welfare from a two-grade tenderness label would be estimated at:
105,000,000 (number of households in the US) × 52 (number of purchase occa-
sions per household, per year) × $0.173 (welfare gain per choice occasion) =
$944.58 million per year under the stated price assumptions. One can draw
broader conclusions about the total welfare effects of a tenderness grading sys-
tem by comparing these estimated benefits against the costs of implementing
and maintaining such a grading system.
1 For example, Buzby et al. (1996) estimate that for six bacterial pathogens, the costs of human ill-
ness are estimated to be $9.3–$12.9 billion annually. Of these costs, $2.9–$6.7 billion is attributed
to food-borne bacteria.
122 Experimental Auctions
an estimated 1.5 billion annual episodes of diarrhea which result in more than
three million premature deaths” (Brundtland, 2001).
Van Ravenswaay (1988) emphasized that the key economic question in food
safety research is to determine individual willingness to pay for reduced risk.
She argues that “we know nothing about the demand for food safety . . .” and
that more research is needed to develop methods to evaluate willingness to pay
values. In response, Hayes et al. (1995) developed an experimental auction to
elicit the value for safer food. Their experimental markets elicited both option
price (i.e., WTP) and compensation (i.e., WTA) measures of value for five food-
borne pathogens. They also constructed six treatments to evaluate how subjects
responded to changes in the risk of illness for a given pathogen, Salmonella, to
explore if pathogen-specific values act as surrogate measures of general food
safety preferences. We now review their objectives, experimental design, and
findings for policy.
What do you think is the chance of becoming ill from (pathogen name), given that
you eat an average amount of typical products in the United States over one year?
Answer: – chance out of 1 million people.
Second, they measured the ex ante economic value of reducing health risks
posed by the pathogens. They estimated either the option price (R) for a lower
probability of illness on a per meal basis, or the compensation demanded (C) for
a greater probability of illness. Let a person’s state-dependent preferences be
defined by a utility function, U(M,H), where M is monetary wealth and H reflects
the person’s health state (H = 0 if sick; H = 1 if healthy), U S (M) ≡ U(M, 0)
and UH (M) ≡ U (M, 1), where U H (M) > U S (M). Assume Ui > 0 and Ui < 0
(i = S, H ), where primes denote first- or second-order derivatives. If a person
has a probability pi of getting sick from pathogen i and (1 − pi ) chance of
remaining healthy, their expected utility is:
EU i = pi U S (M) + (1 − pi )U H (M) i = 1, 2, . . . 5.
Valuation case studies 123
The maximum the person will pay to eliminate the probability of illness
is Ri :
U H (M − Ri ) = EUi i = 1, 2, . . . 5.
The minimum compensation, Ci , they would accept to increase the chance of
illness from pathogen i, is:
U H (M) = pi U S (M + Ci ) + (1 − pi )U H (M + Ci ) i = 1, 2, . . . 5.
Third, Hayes et al. (1995) examined how people respond to increases in the
probability of illness holding constant the illness severity for Salmonella. They
created an environment to help ensure that people would believe the chances
by conducting the experiments in a building where animals are slaughtered and
processed, new food products are continually being made and tested, and large
numbers of bacterial plate counts are conducted on a daily basis. The building
adjoins a meat irradiation facility where pathogens on meat can be eliminated
via high-energy gamma rays. From a Bayesian perspective, a person’s risk
α p
perception is represented by a general beta distribution p̂i = iαii ++ββi q , where
i
q represents the new information on the chance of illness for Salmonella, and
α i and β i are the weights he or she assigns to his or her prior beliefs, pi , and
the new information. If the chance of illness supplied by the new information is
less than his or her prior belief (q < pi ), the perceived risk will exceed the new
information ( p̂i > q); the chance of illness will be overestimated. If q > pi , the
opposite occurs, p̂i < q. The question is how bidding behavior is affected by
changes in the probability of illness. Totally differentiating the above equation
and rearranging yields:
dRi βi U H − US d 2 Ri
= > 0 and = 0.
dq αi + βi UH dq 2
The option price increases as the probability of illness increases, but it is
conditioned by the person’s relative weight on the new information and his or
her prior beliefs. If the new information is not trusted (β i = 0), the option price
remains unchanged. If priors receive some weight, option price increases as the
chances increase.
The fourth purpose of the research conducted by Hayes et al. (1995) was to
examine whether bids for specific pathogens acted as surrogate measures for
general food safety preferences. Surrogate bidding occurs when prior beliefs
on odds seem to motivate bidding more than the new information on the objec-
tive odds, that is, illness option prices are relatively flat across the proba-
bilities of illness. They tested for surrogate bidding by comparing the value
estimates from each pathogen treatment with the value estimate from a treat-
ment that combines all the pathogens. The objective risk of the combined
124 Experimental Auctions
pathogens is a 1 in 46,585 chance of illness per meal from at least one of the
pathogens.
Step 7 introduced new information after the tenth round on the two key
elements that define risk: the objective probability and health severity of the
pathogen. Each subject received only one pathogen description, not all five.
For example, the information given in the treatment for campylobacter was as
follows:
r Test product: If you eat this food, there is a 1 in 125,143 chance that you will
become ill from campylobacter.
r Stringently screened: This food has been subjected to stringent screening for
Campylobacter. There is a 1 in 100,000,000 chance of getting Campylobac-
teriosis from consuming this food.
r Symptoms are those of an intestinal disease with acute diarrhea and severe
abdominal pains. Diarrhea is preceded by brief fever and malaise. The actual
individual chance of infection of campylobacteriosis is 1 in 114 annually. Of
those individuals who get sick, 1 individual out of 1,000 will die annually.
The average cost for medical expenses and productivity losses from a mild
case of campylobacteriosis is $230.
Step 8 elicited informed bids over ten additional rounds. In all, each subject
made twenty bids: ten uninformed and ten informed. Step 9 determined which
of the twenty trials would be randomly selected as binding. We used one bind-
ing trial to control for wealth effects. Step 10 required all subjects to eat their
sandwich to reinforce the non-hypothetical nature of the experimental auction.
In the option price treatments, only the winner of the binding trial ate the strin-
gently controlled sandwich. In the compensation treatments, all but the lowest
bidder ate the stringently controlled sandwich. Each subject was required to eat
the sandwich (either a test product or the stringently screened sandwich) before
receiving money. Each subject signed a second agreement of understanding
before eating the food.
Subjects were recruited from several non-intersecting undergraduate classes
at Iowa State University by announcing the chance to participate in an exper-
iment providing an approximate stipend of $18 and a “free lunch”. Fifteen
participants and two alternates were chosen from each class and were asked
to appear at a specified date and time at an on-campus taste-testing room with
modern kitchen facilities. The taste-testing room is regularly used to measure
reactions to experimental products developed at a nearby facility. A total of 230
subjects participated in the experiment.
For each treatment, subjects were given an identification number and asked
not to communicate with each other. Each subject was given a set of instructions,
with the only differences across treatment being the name of the pathogen and
the trial 11 description of the probability and severity of illness. After reading
the instructions, the subjects answered a set of questions to determine their
understanding of the experimental auction market. After the monitor answered
all relevant questions, the experiment began.
126 Experimental Auctions
7.3.3 Results
Four main results emerged. First, subjects underestimated the objective risk
of food-borne pathogens. As shown in Table 7.4, the average subject overesti-
mated the likelihood of only one of the five pathogens – Clostridium, the lowest
probability pathogen. More than 96% of the subjects underestimated the risk
of Salmonella, the most common and familiar of all pathogens. For Campy-
lobacter, with the highest probability of illness, nearly 94% underestimated the
risk. One explanation is that people sometimes do not understand the nature of
food-borne illness since it displays symptoms similar to those of the flu.
Second, value measures across food-borne pathogens were not robust to
changes in the relative probabilities and severity, suggesting that people place
more weight on their own prior perceptions than on new information on the
odds of illness. Given market participation and full information, one can make
the case that the bids in trials 17 to 20 are the most useful for policy. The bids
reflect information obtained after the bidding process had stabilized with market
experience. The bids suggest the typical participant was willing to pay between
$0.42 and $0.86 per meal to reduce the objective probability of food-borne
illness caused by the presence of each pathogen to a 1 in 100 million chance.
Table 7.4 also compares the mean naı̈ve bids of trials 7 to 10 with the mean
informed bids of trials 17 to 20. The t-test and signed-rank test indicate that the
option price differences between naı̈ve and informed bids for Salmonella and
Clostridium perfringens were statistically significant at the 1% level. The mean
differences for Campylobacter and Staphylococcus aureus were significant at
the 5% level, and the mean differences for Trichinella spiralis were significant
at the 10% level. Overall, full information and repeated exposure to the auction
market impacted average option price values.
Third, marginal willingness to pay decreased as probabilities increased, sug-
gesting people put more weight on their prior beliefs even as new information
suggested greater chances of illness (see Table 7.5). The additional Salmonella
treatments that varied only the probability of illness shed some light on the orig-
inal experiments. Results indicate that, given the small sample sizes, attempts
to rank the pathogens by using the compensation or option price trials would
be meaningless. The intra-group variability (as measured by the range of the
salmonella results) is greater than the variability of responses among pathogens.
Any differences among the pathogens in the first ten trials therefore can be
attributed to the different participant groups we used. These results indicate
that participants overestimated the smallest odds (1 in 1.37 million), under-
estimated the largest odds (1 in 13.7), and reacted to the presence of risk as
much as to the actual level of risk. These results are inconsistent with the
argument that option price increases linearly with increases in perceived risk.
Instead, the explicit Bayesian updating process appears not to be linear. Instead
Valuation case studies 127
Table 7.4. Subjective and objective risk and a comparison of naı̈ve and
informed option price (Ra ) of five pathogens
Naive and
Annual probability of informal option Comparison of naive and informal
Treatment illness price ($) option prices ($)
H0 : R7−20 = R7−10
H1 : R17−20 = R7−10
R17−20 = 0.86
(0.38)c Riff = 0.15
Campylobacter 1/114 1/910 R7−10 = 0.71 (0.30)c 2.33* 24*
(0.43)
R17−20 = 0.55
(0.25) RDiff = 0.11
Salmonella 1/125 1/194 R7−10 = 0.44 (0.10) 4.13** 39**
(0.23)
R17−20 = 0.84
(0.33)
Staphylococcus 1/159 1/2,681 R7−10 = 0.92 RDiff = −0.08 −2.14* −21*
aureus (0.32) (0.12)
R17–20 = 0.81
(0.55) RDiff = 0.12
Trichinella 1/2,400 1/5,665 R7−10 = 0.69 (0.25) 1.59 19
spiralis (0.46)
R17−20 = 0.42
(0.33)
Clostridium 1/24,000 1/287 R7–10 = 0.58 RDiff = −0.16 −2.25* −26**
perfringens (0.41) (0.27)
H0 : Ra 17–20th = R7–10th
H1 : R17–20th = R7–10th
R17–20 = 1.42
(0.57) f RDiff = 0.88
1/13.7 R7–10 = 0.54 (0.36)c 9.04** 52.5**
(0.30)
R17–20 = 1.76
(0.80) RDiff = 0.88
1/137 R7–10 = 0.88 (0.57) 5.84** 45.5**
(0.45)
R17–20 = 0.50
(0.21) RDiff = −0.02
1/1,370 R7–10 = 0.52 (0.09) −0.84 −11.5
(0.20)
R17–20 = 0.92
(0.30) RDiff = 0.25
1/13,700 R7–10 = 0.67 (0.12) 8.15** 52.5**
(0.23)
R17–20 = 0.55
(0.25) RDiff = 0.11
1/137,000 R7–10 = 0.44 (0.10) 4.13** 39.0**
(0.23)
R17–20 = 0.02
(0.06) RDiff = −1.30
1/1,370,000 R7–10 = 1.32 (0.93) −5.42** −45.5**
(0.95)
Note: The sample sizes are as follows: 1/13.7 (n = 14), 1/137 (n = 14), 1/1,370
(n = 15), 1/13,700 (n = 15), 1/137,000 (n = 15), 1/1,370,000 (n = 15).
a R = Option price in dollars.
b Mean of trials 17 through 20 and mean of trials 7 through 10.
c Difference between the mean of trials 17 through 20 and mean of trials 7 through 10.
d The t-statistics are shown and ** denotes rejection of H at the 0.01 significance level
0
for the two-tail t-test.
e The critical values are shown and ** denotes rejection of H at the 0.01 significance
0
level for the Wilcoxon signed-rank test.
f Sample standard deviations are in parentheses.
Valuation case studies 129
the observation that subjects respond to the log of the probability and not the
actual probability suggests that the weight they place on the new information is
affected by the new information. The average subject placed increasingly less
weight on the new information the greater the chance of illness.
Fourth, the evidence does not contradict the view that individual values for
specific pathogens might act as surrogates for general food safety preferences.
The information provided in trial 11 on the probability of suffering from at least
one of the five pathogens was the sum of individual pathogen odds, 1 in 46,585.
If surrogate bidding exists, the values elicited for the combination of pathogens
would not be significantly different from the values elicited for each individual
pathogen. The average bid at trial 10 was $0.73, and the average bid at trial
20 was $0.78. The results do not reflect the increase in probability of illness,
again suggesting that consumers responded to the presence of pathogens in food
rather than to the levels of individual pathogens.
7.3.4 Conclusions
In conclusion, the general policy implications suggest the average subject in
our experimental environment was willing to pay approximately $0.70 per meal
for safer food. The Salmonella treatments under alternative probability levels
indicate that the average subject would pay approximately $0.30 per meal to
reduce risk of food-borne pathogens by a factor of 10. If one could transfer these
values to the US population, the value of food safety could be at least three times
the largest previously available estimates. But one should also recognize these
values represent upper bounds on value. The La Chatelier principle suggests
these current option price bids represent a measure of the upper end of the
distribution of food safety preferences, the experimental auctions should be
constructed such that people participate in daily auctions over a week or more
and be provided more substitution possibilities.
7.4.1 Experiment
We now examine the experimental auctions used by Rousu et al. (2004b) to
address how US consumers react to a positive tolerance standard for GM ingre-
dients. They designed an experimental auction using three GM products to test
two hypotheses: (a) mean bids for the GM-free products equal the mean bids
for the GM-threshold products, set either at 1% or 5% and (b) mean bids for the
1% GM product equal the mean bids for the 5% GM product. In their experi-
ment, participants were randomly assigned to a treatment that differed in terms
of the tolerance level. Consumers bid on three food products that had different
tolerance levels. In one trial, all consumers bid on foods with a non-GM label,
Valuation case studies 131
In Step 2 participants were given detailed instructions (both oral and written)
about how the random nth price auction works. A short quiz was given to ensure
everyone understood how the auction worked. In Step 3 the random nth price
auction was introduced by conducting an auction in which the consumers bid on
one brand-name candy bar. Each consumer examined the candy bar, submitted
a bid, and the auction was run for real.
In Step 4 the second practice round of bidding was run, and consumers bid
separately on three different items: the same brand-name candy bar, a deck
of playing cards, and a box of pens. Participants knew only one of the two
rounds would be chosen at random to be binding, which prevented anyone from
taking home more than one unit of any product. Selecting a random binding
round eliminates the threat of a person reducing his bids because he could buy
more than one unit. The consumers first examined the three products and then
submitted their bids. In Step 5 the binding round and the binding nth prices were
revealed to the consumers. All bid prices were written on the blackboard, and
the nth price was circled for each of the three products. Participants could see
the items they won and the market-clearing price. The participants were told
that the exchange of money for goods was in another room nearby and would
take place after the entire experiment was completed.
After Step 5 the GM-food products were introduced for the next two rounds
of bidding. The two bidding rounds were differentiated by the food label – either
a non-GM label certified to be GM-free or a non-GM label that indicated the
tolerance of GM material. Figure 7.1 shows the three types of labels used for
the vegetable oil product; the other product labels were constructed similarly.
These labels were on the front of the package and large enough for participants
to easily read them. In one round (which could be round one or two depending
on the experimental treatment), participants bid on the three food products each
with the certified non-GM food label. In the other round, participants bid on the
same three food products with the 1% or 5% GM tolerance level. Consumers
knew that only one round would be chosen as the binding round that determined
auction winners.
In Step 6 consumers submitted sealed bids for the vegetable oil, tortilla chips,
and potatoes, either with the certified non-GM label or the GM-tolerant label.
Each consumer bid on each good separately. The monitor collected the bids
and then told the participants that they would now look at another group of
food items. In Step 7 consumers examined the same three food products, each
with a different label from round 1. Again they examined the products and
submitted their bids. Consumers bid on food products with only two types of
labels, the GM-free and the GM-tolerant label. Consumers either saw the 1
or 5% tolerant labels but not both. In Step 8 the monitor selected the binding
round and determined the random nth prices for the three goods and notified
the winners. A participant might purchase a bottle of vegetable oil and a bag
Valuation case studies 133
Figure 7.1 The three types of labels used for the vegetable oil
134 Experimental Auctions
of potatoes, for example, but there was no chance they would win two bottles
of vegetable oil. The assumption here is that vegetable oil, potatoes, and chips
are unrelated in consumption. In Step 9 each consumer completed a brief post-
auction questionnaire, and the monitors dismissed the participants who did not
win. The monitors and the winners then exchanged money for goods, and the
auction ended.
Rousu et al. (2004b) used several design features that differed from previous
studies. First, subjects submitted only one bid per product. They stepped back
from the protocol of using multiple repeated trials and posted market-clearing
prices to avoid any question of creating affiliated values that can affect the
demand-revealing nature of a laboratory auction. Second, subjects were not
endowed with any food item and were not asked to “upgrade” to another food
item; rather participants were paid $40 and then they bid on different foods in
only two trials. This avoids the risk that an in-kind endowment effect distorts the
participant’s bidding behavior, but introduces the potential for bids for goods
to be influenced by the price of oil, potatoes, and chips outside the experiment.
Third, each consumer bid on three food items assumed to be unrelated in con-
sumption, such that if he or she did not have positive demand for one or two
products, information on their tastes for genetic modification was obtained on
the second and (or) third products. Finally, as did Lusk et al. (2004a) they used
adult consumers over eighteen years of age who were chosen using a random
digital dialing method.
7.4.2 Results
The demographic characteristics of the sample, while not perfectly matching
the population demographic characteristics for this region as reported by the
US Census Bureau, were similar and provided a sufficient representation for
the initial probe into labeling and information for GM products. This is par-
ticularly true recognizing that adults, rather than students, are likely to better
reflect a typical household of consumers. Using a national random sample of
grocery store shoppers, Katsaras et al. (2001) showed that the share of college-
age (eighteen to twenty-four years) shoppers falls far below their share in the
population – 8.5% of shoppers versus 12.8% in the US Census of Population.
College students obtain a large share of their food from school cafeterias and a
small share from grocery stores and supermarkets compared to older shoppers
(Carlson et al. 1998). Although our participants are slightly skewed toward
women, Katsara et al. (2001) show that women make up a disproportional
share of grocery shoppers – 83% of shoppers versus 52% in the US Census of
Population.
Two primary results emerge. First, consumers bid less for the products having
GM-tolerance levels relative to the GM-free benchmark. Table 7.6 shows the
Valuation case studies 135
B. Mean bids when participants bid on food with a 5 percent tolerance level
C. Mean bids when participants bid on food with a 1 percent tolerance level
mean and median bids by food type. Twenty-eight participants bid in the 5%
tolerance treatments; sixteen participants bid in the 1% treatment. Overall, the
average consumer bid less on the food product with the GM-tolerance labels
relative to the GM-free products. Consumers, on average, bid seven cents less
on the GM-tolerant oil, fourteen cents less on the tortilla chips, and nine cents
less on the potatoes. Consumers, on average, discounted the foods with the GM
tolerance by an average of seven to thirteen percent. In comparison, Rousu et al.
(2004a) observe that consumers discounted food that had a GM label without
a tolerance level by an average of fourteen percent. Pooling all observations,
Table 7.7 shows we can reject the null hypothesis that bidding behavior over
GM-tolerance labels is identical to that for the GM-free benchmark for the
tortilla chips and the potatoes but not for the vegetable oil. The significant
discount for the GM-tolerant food is consistent with Viscusi et al. (1987) which
indicate a reference risk effect. In their study, consumers initially purchased a
given product when told that it injured fifteen out of 10,000 people who used
136 Experimental Auctions
the product, but over two-thirds of the consumers were unwilling to purchase
the same product when the chance of injury increased to sixteen out of 10,000.
Such findings could explain why consumers placed such a large discount on
the GM-tolerant food.
Second, no statistically significant difference existed for consumers’ discount
of the 5% GM products and 1% GM food. Table 7.7 shows that at the 5%
significance level we cannot reject the null hypothesis that bid differences are
different between the two GM-tolerance levels.
7.4.3 Conclusions
The findings in this case study support the view that if a GM-tolerance policy
is implemented in the US, consumers might not place a greater value on a
1% GM tolerance level relative to a 5% GM tolerance level. Because of the
higher segregation and handling cost of a 1% tolerance level compared to a 5%
level, society may be better off implementing a higher tolerance level, if it is
in fact optimal to implement a mandatory label at all (see Lusk et al., 2005).
Consumers value GM-free products, but if GM contamination does exist, we
Valuation case studies 137
Non-GM Non-GM
premium – 5% premium – 1% Difference T-Test statistic
where WTPij is the maximum willingness to pay for brand j (i.e., the bid for
j in an incentive compatible auction) and pj is the price of good j. There are
several ways of utilizing this utility function to determine market share. One
common approach is to use the “first choice” or “highest utility” rule in which
each person is assumed to choose the product producing the highest utility from
a choice set with J brands. Assume individual i chooses brand j if Uij > Uik for
all k = j. Let Iij be an indicator variable that takes the value of 1 if individual i
chooses brand j and 0 otherwise. Further, let wi be the number of units individual
i purchases once a choice between brands has been made. Now, in a sample of
N individuals choosing between J goods, the market share of brand j (MSj ) is:
N
Iij wi
i=1
MS j = . (7.13)
J N
Iik wi
k=1 i=1
on substitution patterns. Further, the MNL and other such models often assume
homogeneity in preferences across people. Although a variety of econometric
models have been proposed to undercover heterogeneity in choice and market
share data, auction bids provide direct, continuous measures of value for each
person for each good, which is the most one can ask for when attempting to
identify heterogeneous preferences.
7.5.2 Results
Table 7.9 contains the results of the market share simulations. In the base-line
scenario, about 14% of subjects were predicted to choose the generic steak,
whereas 27% and 29% of consumers were predicted to choose the Choice and
CAB steak. At the price levels used in the simulation, about 30% were not
140 Experimental Auctions
Scenarios
Market share
Generic 19.41% 22.14% 19.41%
Choice 47.26% 62.01% 46.37%
CAB 33.33% 15.86% 32.15%
Natural 0.00% 0.00% 2.07%
expected to choose a steak at all; the prices of all steaks exceeded the bids for
all steaks for 30% of bidders in the sample. It is not unrealistic to suppose a
similar phenomenon occurs on routine shopping trips as consumers only buy
if the price is right. In terms of market share in the baseline scenario, almost
half the share went to Choice steak, a third went to CAB, and the remainder to
generic. The second column of Table 7.9 illustrates the effect of a $0.50 CAB
price increase. As expected, demand for CAB fell when the price increased;
market share of CAB declined about 17%, indicating an own-price elasticitiy of
(−17.47/15.86)/(0.5/4) = −8.81. The Choice steak experienced a much larger
increase in market share due to the CAB price increase than did generic, indi-
cating Choice and CAB are stronger substitutes. The cross-price elasticities of
Choice and generic with respect to CAB are 1.90 and 0.98, respectively. The
last column of Table 7.9 shows the effects of introducing a new steak to this
Valuation case studies 141
would provide at least twelve chops of each treatment (i.e., all twelve chops
needed for each treatment were from the same carcass). Two days before the trial
excess back-fat was removed to a common 1/8–1/4 inch depth, corresponding
to current retail marketing standards, and the loins were cut into chops of 3/4 to
one inch thickness. The chops were individually vacuum wrapped, marked for
identification, and stored in the meat cooler. Despite the extensive search, loins
corresponding to two of the extreme treatments (5-5-3 and 5-1-3) could not be
located due to their low frequency in nature. The fresh products used for visual
and taste assessments of attributes were limited to the 1-1-3 chop in each repli-
cation. On the morning of the trial, two chops of each treatment were removed
from their vacuum packaging, placed on a yellow Styrofoam tray, and wrapped
with clear plastic in a manner similar to local retail marketing standards. The
eight packages representing the alternative treatments were marked for identi-
fication and were randomly placed in a commercial chest-type retail cooler at
45◦ F in two rows of four chops each. To these we added the T-bone steak and
skinless chicken breast purchased at retail on the morning of the trial. The cooler
was covered to avoid incidental and premature viewing. Both photographs and
fresh chops were coded for experimental identification so consumers could not
discern the independently determined level of treatment variables represented
by the chop (i.e., its level of color, marbling, or size). Furthermore, they could
not associate the photographs with the fresh product by treatment identification
code.
The multi-unit auction was conducted in five phases. Phase I: the consumers
were re-surveyed to update information on frequency of pork chop consumption
and current stock of pork chops. Phase II: consumers learned about how the
auction worked by a pre-auction of eight candy bars. A second price Vickrey
auction format was employed.
Phase III: consumers were shown photographs of the chops and asked to eval-
uate the chops’ attributes of color, marbling, size, bone, and overall appearance
using a scale from “Strongly Dislike” to “Strongly Like.” Consumers were then
provided reference prices for the beef T-bone ($5.00 per pound) and the chicken
breast ($3.00 per pound) and asked to simultaneously submit hypothetical bids
for each of the eight chops represented by the photographs. These bids were
described as hypothetical because consumers were aware no purchase would
occur (in that phase).
Phase IV: consumers were shown the actual chops, including the beef T-bone
and chicken breast, and were asked to evaluate their appearance using forms
identical to those used in Phase III. The reference prices for the beef T-bone
and chicken breast were the same as in phase three, and the consumers were
asked to submit eight simultaneous bids in a second-price auction. Consumers
were told the market price (second high bid) and the identification number of
the highest bidder for each of the eight pork chops. Consumers were given the
Valuation case studies 145
opportunity to increase their bid for any of the eight chops and the bidding was
continued in this manner until no one wanted to increase their bid for any chop.
Phase V: consumers were escorted to another room where they tasted a sam-
ple (approximately 3/4 inch cube) of a professionally prepared chop broiled to
71◦ C internal temperature from ribs adjacent to the eight fresh chops viewed
in phase four. Participants were asked to evaluate each chop tasted for fla-
vor, juiciness, tenderness, and overall eatability using a scale comparable to
their previous visual evaluations. After completing the taste test, consumers
then returned to the original auction room and were allowed to re-inspect the
fresh chops using, if desired, their taste evaluation forms as added information.
Consumers were then asked to re-bid for fresh chops. As in Phase IV, the auc-
tion ended when no one wished to increase his/her bid for any chop. Each of
the eight chops was sold to the highest bidder in packages of two at the second-
highest bid per pound for that chop. The chops not purchased by the high bidder
were then sold to the second highest bidder at the third highest price bid for
that chop, which became the new market price. Note: consumers were kept
unaware of any subsequent phase of the experiment to enhance the validity and
independence of data collected within each phase. Anonymous consumer iden-
tification numbers were changed between Phases IV and V to further enhance
the independence of data collected.
small amounts of marbling and light to pale white color (Table 7.10). But
after tasting cooked samples, consumers exhibited a preference for chops hav-
ing greater marbling and to some degree for uncooked chops having darker
color and smaller size than those chosen based on photographs. Rankings after
taste-testing tended to contradict consumers’ preferences obtained from visual
appraisals. As a result, the correlations of overall chop eatability, as measured on
a scale, with overall chop appearance rankings were negative for photographs
and small, but positive, for fresh product.
Melton et al. (1996) also examined price differences associated with subjec-
tive evaluations. They considered both bids (one per consumer per chop) and
market prices (one per replication per chop). The bids provide an indication of
each consumer’s willingness to pay for a fixed supply, whether or not a sale
is actually consummated at that price. The experimental market price approxi-
mates the competitive market price that could be observed in a non-experimental
setting in that a number of consumers have been “priced” out of the market and
are unobserved. Table 7.11 reports the results of several ordinary least squares
regressions that summarize the effect of color, marbling, size, and presentation
format on market prices and individual bids. Overall, results tended to follow
the pattern of overall preference ratings. Focusing on market prices, results
indicate higher prices for lower marbled and larger cuts in the photo evaluation,
but higher prices were observed for higher marbled cuts in the taste evaluation.
Individual bids exhibit a similar pattern: higher bids are offered for larger, less
marbled, and lighter color chops when evaluated visually. After tasting, how-
ever, consumers tended to bid more for more marbling whereas color and size
had very little effect on bids.
Although appearance is an important consideration in consumers’ fresh pork
purchases, repeat purchases are likely to be most affected by taste-assessed
attributes. Our results suggest that first-time purchasers of pork chops may
inadvertently be selecting, based on appearance, what becomes their least pre-
ferred tasting pork quality, or at least a product that is likely to be judged as
less desirable. We cannot easily identify the source of these apparently incon-
sistent and contradictory preferences. However, we suggest that the observed
preferences for less marbling and lighter color in fresh pork appearance may
be a result of the growing health concern for reduced dietary fat intake and
the “Other White Meat®” advertising campaign that promotes pork that looks
good but tastes bad. Such strategies seem unlikely to sustain long-term market
success.
Consumers may be over-reacting to the “low fat” health concerns when
making visual appraisals. Actual fat differences between marbling score 2 and
marbling score 4 are small (about 2%), but taste differences appear to be much
greater. This inconsistency suggests that consumers may not be well informed
about the taste-health tradeoffs of intra-muscular (i.e., marbling) as opposed
Table 7.10. Means of subjective evaluation scores of pork chop characteristics and auction bids by presentation format
(Scale = 1 to 100)
Chop Color Marb. Size Overall Color Marb. Size Overall Juiciness Tender Eatability
(C-M-S) score score score score Bid score score score score Bid score score score Bid
2-2-3 37.5 50.3 58.6 46.3 $1.97 41.0 54.5 60.9 56.5 $1.94 32.2 30.4 35.2 $1.49
2-4-3 40.0 41.5 58.6 48.4 $2.03 50.8 40.7 48.6 50.5 $1.95 55.6 60.7 61.0 $2.10
3-3-2 48.1 36.8 28.8 28.8 $1.72 41.0 52.6 29.7 35.7 $1.62 40.8 53.6 48.9 $1.88
3-3-3 58.9 52.7 53.3 53.4 $2.22 41.9 41.1 52.1 42.6 $1.69 42.8 37.0 40.1 $1.59
3-3-4 53.4 61.0 68.5 59.8 $2.17 55.5 48.8 60.4 54.2 $1.95 40.6 44.3 44.3 $1.66
4-2-3 65.1 66.5 47.1 63.7 $2.36 37.2 48.0 36.9 39.0 $1.79 42.3 36.9 42.5 $1.71
4-4-3 51.3 41.2 41.7 35.4 $1.94 37.3 29.1 42.3 38.8 $1.76 61.9 58.8 57.8 $2.10
1-1-3 35.1 38.6 52.8 47.1 $2.13 51.4 66.6 64.0 62.5 $2.08 39.9 40.9 44.2 $1.71
5-1-3 51.2 62.4 39.6 50.9 $1.97
5-5-3 40.3 11.7 18.1 15.2 $1.53
148 Experimental Auctions
a Bid regressions also included nineteen other variables related to socio-economic and
demographic characteristics as well as pork consumption patterns.
b Numbers in parentheses are t-statistics.
c These dummy variables are effects coded: the estimated effects across the three repli-
cations sum to zero and are estimated relative to the mean replication effect.
7.6.3 Conclusions
Melton et al.’s (1996) experimental auction was designed to evaluate consumer
perceptions and willingness to pay decisions for fresh pork chops that dif-
fer in selected attributes: color, marbling, and size. They found that consumers
attached value to the attributes of color, marbling and size in fresh pork; appear-
ance and taste effect consumer purchase decisions differently; consumers did
not consistently prefer white chops over dark red chops, but instead varied their
preference depending upon presentation format. The results showed evaluations
based on taste were positively related to visual evaluations of the fresh chop,
but consumer evaluations of chops based on photographic format were shown
to be unreliable sources of market information.
2 We would like to thank Christine Wilson and Dana Marcellino for permitting use of the infor-
mation used in this case study.
150 Experimental Auctions
Although the academic literature is replete with examples exposing the merits
of farm record keeping and investigating the link between record keeping and
financial performance, very little is known regarding the value farmers place on
their financial information and the determinants of such value. Such information
is needed as public institutions determine the quantity and quality of resources
to devote to encouraging and supporting maintenance of farm records.
identifying the type and quality of records they possessed. For example, subjects
were asked whether they maintained a balance sheet, statement of cash flows,
income statement, statement of owner’s equity, checkbook register, tax records,
etc. For each item listed, farmers were asked if they prepared the item listed,
for how many of the past five years had they prepared the item, and if the item’s
form was handwritten or electronic.
After the participants filled out the financial records inventory sheet, they
were requested to decide on an amount, e.g., a bid, for which they were willing
to sell all the documents listed on their inventory. Because the bid consists
of the price at which participants were willing to sell several different types
of financial information, the participants were asked to indicate, in percentage
terms, the amount that reflected their value for the balance sheet, statement of
cash flows, income statement, statement of owner’s equity, etc.
Several key points were emphasized to the participants. They were informed
that bids would be collected at several locations over a time period of sev-
eral days. The lowest bidder across all locations would be contacted later to
be informed if they had won. It was also stressed that the auction was not
hypothetical; the winner would receive real money for his/her financial records.
The instructions emphasized that the winner was expected to give all originals
and copies of the records listed on their individual financial records inventory
sheet. The monitor made it clear that the farmer with the lowest bid would
receive in cash the overall second lowest bid price, but he/she would forfeit
his records. A tax audit was the only exception to the winner of the auction
regaining the right to view his/her records. As in the candy bar auction, it was
explained, in detail, why the best strategy was to submit a bid exactly equal to
the amount that would make the person indifferent between money and their
records. It was explained that no bid was too small or too high. The partici-
pants were also told that if there was no amount of money they were willing to
accept for their financial records, they could select that option on the bid sheet.
An opt-out option was offered to discourage participants from writing down
an artificially high bid price in “protest” and to discourage participants from
leaving the bid sheet blank. Instructions are included in Appendix.
7.7.2 Results
Of the 72 participants in the study, 53 submitted positive bids, 18 people checked
the box on the bid sheet indicating that there was no amount of money they
would accept for their records, and one person did not complete their bid sheet.
Of those people submitting bids, the range was from $100 to $2,500,000, with
the average bid being $145,657 and the median being $25,575.
Because of the wide dispersion of auction bids, farmers were placed in one of
four categories: bids between $0 and $14,999, between $15,000 and $75,499,
152 Experimental Auctions
and $75,000 and greater, and farmers who did not submit a bid. Table 7.12
shows the means for several record, farm, and farmer characteristics segregated
by the four bid categories. Results indicate those people submitting the lowest
bids spent the least amount of time, on average, preparing their records. The
hypothesis that the mean time spent preparing records is the same for each
of the four bid categories can be rejected at the p = 0.01 level according to
an ANOVA test. Time spent examining records was roughly increasing in the
amount bid for financial records, but the means were not statistically different
across category. Table 7.12 shows a pronounced difference in the value of farm
records across farm size. Farmers in the lowest bid category ($0 to $14,999)
had gross farm sales 4.26 times lower ($1,015,000 vs. $238,160), on average,
than farmers in the highest bid category ($75,000 and greater).
The results in Table 7.12 provide some insight into the characteristics of
people choosing not to bid in the auction. Such people tended to be the least
educated, the oldest, and while having relatively large farms in terms of acreage,
their gross sales were relatively low (i.e., they were low performing as exhibited
by relatively small gross sales per acre).
Each person’s bid included their value for six potential types of records (bal-
ance sheet, statements of cash flows, income statement, statements of owner’s
equity, checkbook register, and tax records). After submitting their bid, farmers
indicated, of the total bid amount, the percentage value attributable to each
record type. By multiplying the percentages by the bid, a value for each record
type can be found. For those 51 farmers who submitted a bid, results indi-
cate that tax records, on average, were the most valuable record type with a
calculated value of $39,755, followed by the balance sheet, which was worth
$30,747. On average, the two least valuable types of records were the statement
of cash flows and the checkbook register.
7.7.3 Conclusions
This case study illustrated how experimental auctions can be used to elicit the
values for an extremely high-valued good: business records. Results indicate
that, for the sample of farmers considered, farm records were extremely valu-
able. On average, people bid $145,657 to give up their farm financial records.
Results also suggest wide diversity in valuations with bids ranging from $100
to $2,500,000. Some of this diversity is explained by the characteristics of
the records and farmers: farmers who owned large terms tended to value their
records more as did farmers who spent more time preparing their records.
This case study also demonstrates some challenges in implementing experi-
mental auctions for high-valued goods. For this study, the theoretical construct
of interest was willingness to accept. We were interested in the value people
placed on the records they owned. The study indicated that a quarter of the
Table 7.12. Distribution of bids for farm records and characteristics of farmers in four bid ranges
a 1 = high school credit, 2 = high school graduate, 3 = college credit, 4 = college graduate, 5 = graduate credit, 6 = graduate degree.
b P-value from an ANOVA test of the hypothesis that the mean value of the variable in the column is the same for all bid ranges.
154 Experimental Auctions
bidders were not willing to sell their records. Such a finding is difficult to
reconcile in that there is likely some amount of money that would make a
person indifferent to having their records and not having them. What the result
likely implies is that some people either do not understand the mechanism
or are “protesting” the auction as conducted. That low-performing, older, and
less-educated farmers were less likely to submit a bid is consistent with this
hypothesis.
This case study illustrates that experimental auction methods need not be
limited to low-valued goods or food. Despite the relatively high frequency of
no-bidders, bidding behavior conformed to a priori expectations, i.e., farms
with higher gross sales valued their records more highly, suggesting a rea-
sonable degree of validity. This case study also demonstrated how experi-
mental auctions can be used to value a relatively complex good comprised
of many sub-components and offered one method for separately valuing each
sub-component.
Virtually all major food retailers in the EU have stopped carrying GM products,
essentially banning GM food from the European market.
The EU contends that differences in consumer concerns for food safety across
the Atlantic justify the precautionary policies regarding GM food. But it is
not readily apparent that EU consumers are more concerned about GM foods
than are US consumers. US agricultural producer groups argue that the more
restrictive labeling policies, moratoriums on approving new GM crops, and
grocery store bans on GM food are simply a way for the EU to protect domestic
agricultural producers from international competition. The US recently filed
a complaint with the World Trade Organization over the EU’s stance on GM
foods. Nevertheless, EU consumers might truly have different preferences for
GM foods than consumers in the US. If so, the interesting issue is why these
differences exist and how these differences should be addressed in international
trade negotiations.
Although some studies have compared EU and US consumer preferences
for GM food, the analysis has relied on hypothetical surveys that asked simple
scale questions or elicited hypothetical choices (e.g., Gaskell et al.). To fully
document whether differences exist, what is needed is a study involving real
financial incentives and economic commitments.
7.8.1 Experiment
We draw from the data reported in Lusk et al. (2004a, 2005, and 2006). In their
studies, an experimental market was constructed and they elicited the mini-
mum amount of compensation randomly recruited female shoppers demanded
to consume a GM rather than a non-GM food. Experimental sessions were
held with over 160 individuals in three diverse US locations: Jacksonville, FL,
Long Beach, CA, and Lubbock, TX; and with over 200 individuals in two EU
locations: Reading, England and Grenoble, France.
In each of the five locations, market research companies were hired to ran-
domly recruit individuals for participation in a “food preference study” by
offering $50 to attend a one-hour research session. Several sessions were held
in each location and about twenty people were assigned to a session. The US
locations were selected to provide diversity in geography, population, and cul-
ture. The selected EU locations are roughly similar to Lubbock, TX in that they
were towns of relatively similar population, are located in agricultural areas,
and possess relatively large universities. Recruitment was restricted to females
because we were interested in food preferences and females are, overwhelm-
ingly, the primary grocery shoppers in most households (see Katsara et al.,
2001). Limiting the sample to females creates a more homogeneous sample
across the geographic locations allowing for a stronger test of the location
effect. Age restrictions were implemented to ensure that a disproportionate
156 Experimental Auctions
Fourth, the fifth price auction was used for several reasons: a) it is theoreti-
cally incentive compatible; b) the market price is endogenously determined and
subjects could incorporate feedback from the market with this mechanism; and
c) it should engage both people with high and low values as a non-trivial num-
ber of people will “win” the auction. Finally, the experiment involved people
bidding to exchange their endowed non-GM cookie for a GM cookie over ten
bidding rounds with the market price being posted at the end of each round (one
round was randomly selected at the end as binding). Given that most partici-
pants would be unfamiliar with GM food, it was felt that repeated rounds were
needed to promote learning about the mechanism and the good. This approach
also permitted an investigation into the effect of information on valuations by
comparing naı̈ve bids in the first five bidding rounds to bids in last five rounds,
where different information shocks were introduced (see Lusk et al., 2004a).
The experimental auction proceeded as follows:
Step 1. Participants were given a free chocolate chip cookie (referred
to as a biscuit in the EU). The cookie was in a transparent package
containing a label that clearly indicated the cookie contained no GM
ingredients.
Step 2. Participants were shown an otherwise identical cookie that was
labelled as being made with GM ingredients.
Step 3. Subjects wrote on a bid sheet the least amount of money they
were willing to accept to exchange their cookie for the cookie con-
taining GM ingredients.
Step 4. Once all bids were recorded, the monitor collected the bid
sheets.
Step 5. The ID numbers of the four lowest bidders were posted in
front of the room along with the fifth lowest bid amount (the market
price).
Step 6. Steps 3 through 5 were repeated for nine additional rounds.
Step 7. At the completion of the tenth round, a random number was
drawn to determine the binding round.
Step 8. Once the binding round was determined, the four winning bid-
ders were paid the fifth lowest bid amount to exchange their cookie
for the cookie made with GM ingredients. All other participants
kept their cookie made without GM ingredients.
Step 9. All subjects ate the cookie they possessed at the end of Step 8.
7.8.2 Results
Table 7.13 reports the summary statistics of the bids in rounds 1 and 5. The
median bid varies greatly by location. In round 5, the median bid from England
is over twice that in any of the US locations and the median bid in France is
158 Experimental Auctions
Location
Round 1
Low bid $0.00 $0.00 $0.00 $0.00 $0.00
20% quantile $0.03 $0.10 $0.05 $0.16 $0.10
Median bid $0.23 $0.50 $0.12 $0.71 $1.57
80% quantile $1.00 $5.00 $0.50 $4.74 $9.80
High bida $100.00 $100.00 $2.00 $158.00 $98.00
Average $2.44 $5.23 $0.33 $4.82 $8.51
Standard deviation $11.91 $16.44 $0.43 $17.53 $17.08
Round 5
Low bid $0.00 $0.00 $0.00 $0.00 $0.00
20% quantile $0.00 $0.05 $0.02 $0.11 $0.10
Median bid $0.10 $0.20 $0.05 $0.40 $1.96
80% quantile $0.50 $1.00 $0.15 $1.58 $11.76
High bida $100.00 $100.00 $2.00 $158.00 $98.00
Average $2.13 $4.03 $0.19 $3.58 $6.95
Standard deviation $11.84 $16.52 $0.41 $17.67 $13.32
a One observation in Florida and two observations in France were in excess of $10,000
and were removed from the analysis.
p 80
e Lubbock, TX USA
r 60
c 40
e
n 20
t 0
p 80
e Long Beach, CA USA
r 60
c 40
e
n 20
t
0
p 80
e Jacksonville, FL USA
r 60
c 40
e
n 20
t 0
p 80
e Reading England
r 60
c 40
e
n 20
t 0
p 80
e Grenoble France
r 60
c 40
e
n 20
t
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Willingness to Accept
($/cookie)
higher in France than in any of the other locations. In France, more than 30%
of people bid more than $5 as compared to only about 8% in England and 2%
to 5% in the US locations. The take home message from the data in Table 7.13
and Figure 7.2 is that consumers in the European locations demanded signifi-
cantly greater levels of compensation to consume the GM cookie than did the
US consumers.
Having established that bid distributions are different across location, the
interesting question is why such differences might exist. In the survey people
filled out before the auction, they were asked a variety of questions that related to
factors hypothesized to influence consumer acceptance of GM foods including:
subjective and objective knowledge of biotechnology, level of trust in informa-
tion about biotechnology from the government and activist groups, perceptions
of the benefits and risks of biotechnology, and general attitudes toward the envi-
ronment, food, and technology. To determine how these variables affect WTA,
quantile regression methods were used (Koenker and Hallock, 2001; Koenker
and Bassett, 1978). As shown in previous tables and figures, there are a few
“large” outliers. As is well known, such outliers have a pronounced influence in
typical conditional mean (e.g., ordinary least squares) regressions. The median
160 Experimental Auctions
Constant 2.443**
(0.328)
Age Age in years −0.013**
(0.004)
Education 1 if obtained university undergraduate degree or higher; −0.014
0 otherwise (0.082)
Income 1 if household income is greater than $50,000/year; 0.003
0 otherwise (0.076)
Homemaker 1 if full time homemaker; 0 otherwise −0.004
(0.109)
Child 1 if children under the age of 16 living at home; 0 otherwise −0.153**
(0.078)
Subjective knowledge Knowledge of facts and issues concerning genetic 0.035
modification in food production (0.025)
(1 = not at all knowledgeable; 9 = extremely
knowledgeable)
Objective knowledge 1 Response to True/False Question: Ordinary fruit does not 0.117
contain genes, but genetically modified fruit does (0.074)
(1 = correct answer; 0 = incorrect answer or don’t know)
Objective knowledge 2 Response to True/False Question: It is impossible to transfer −0.031
animal genes to plants (0.078)
(1 = correct answer; 0 = incorrect answer or don’t know)
Government trust Level of trust in information about genetic modification in −0.021
food production from government agencies such as the (0.018)
USDA and FDA
(1 = strongly distrust; 9 = strongly trust)
Activist trust Level of trust in information about genetic modification in 0.037**
food production from activist groups such as Greenpeace (0.017)
(1 = strongly distrust; 9 = strongly trust)
Risk/benefit In general I believe the use of genetic modification in food −0.086**
production is (0.021)
(1 = risky; 9 = beneficial)
Environment 1 When humans interfere with nature, it often produces −0.013
disastrous consequences (0.019)
(1 = strongly disagree; 9 = strongly agree)
Environment 2 Mankind is severely abusing the environment 0.020
(1 = strongly disagree; 9 = strongly agree) (0.017)
New food I don’t trust new foods 0.037**
(1 = strongly disagree; 9 = strongly agree) (0.017)
Food quality Quality is decisive for me in purchasing foods −0.019
(1 = strongly disagree; 9 = strongly agree) (0.022)
Natural food I usually aim to eat natural food 0.025
(1 = strongly disagree; 9 = strongly agree) (0.019)
Valuation case studies 161
is much less influenced by outliers. As such, Table 7.14 reports the results of
a median regression. In a median regression, parameters are chosen to mini-
mize the sum of absolute errors, which can be contrasted with OLS parameter
estimates which result from minimizing the sum of squared errors.
Reported coefficients can be interpreted as the impact of a one-unit change in
the covariate on median WTA holding other covariates constant. For example,
results indicate that a one-year increase in age is associated with a $0.013 decline
in median WTA. The only other demographic variable to significantly influence
median WTA is whether children are living at home; median WTA of those
with children was $0.15 lower than those without. As expected, perceptions of
benefits and risks of GM food strongly influenced WTA: the lower the level
of perceived risk, the lower the compensation demanded. People that were less
trusting of new foods and less optimistic about technology, demanded more
compensation to consume the GM cookie at the median.
The last rows report the effect of location on WTA, holding constant all
the other factors in the table. Even after controlling for differences in age,
income, trust, risk perceptions, and attitudes toward food, the environment, and
technology, consumers in the US and England had significantly lower median
WTA than French consumers. Estimates also suggest median WTA from the
three US locations is also significantly different than median English WTA. An
162 Experimental Auctions
interesting question that arises is whether, and to what extent, differences in lev-
els of demographic and attitudinal variables explain differences in WTA across
location. Recognizing that the estimated location dummy variables show the
effect of “location” when the levels of age, education, risk/benefit, etc. are held
at the same levels across location, it is clear that the relatively large magnitudes
of the estimated dummy variables imply that there is still a significant degree
of unexplained variability in WTA across location that is not accounted for by
differences in levels of demographics, knowledge, trust, and general attitudes,
as measured in this study. In other words, if the estimated dummy variables were
zero, differences in explanatory variables across location would fully explain
unconditional differences in WTA across location. The regression model, how-
ever, can explain some of the unconditional difference in WTA across location.
Comparing the “raw” unconditional bids from round 5 to the estimated dummy
variables in Table 7.14 provides an indication of the extent to which differences
in independent variables explain the differences in unconditional WTA across
location. For example, the unconditional median bid (which does not hold con-
stant differences in any explanatory variable) in Lubbock TX is $1.86 lower
than that in France. However, the conditional median bid (i.e., the estimated
dummy variable) in TX is only $1.46 lower than in France. Thus, differences in
levels of demographics, knowledge, trust, and general attitudes, as measured in
this study, explain $0.40 (or 21.7%) of the unconditional median difference
in WTA across TX and France. Similarly, differences in explanatory variables
across location explain 16.7% of the unconditional median difference in WTA
across California and France and 20.7% of the unconditional median difference
in WTA across Florida and France.
Although we are only able to explain roughly one-fifth of the difference in
unconditional median WTA across the US and France, we find stronger results
for England. Calculations imply that the model explains 55.0%, 27.5%, and
44.6% of the unconditional median difference in WTA across England and
Texas, California, and Florida, respectively. These findings suggest that while
differences in levels of demographics, trust, attitudes about food and technology,
and perceptions of risks and benefits explain some of the variability in WTA
across location, there is still much to learn about what drives differences in
values for GM food across the US and EU, especially with regard to France.
7.8.3 Conclusions
Some US producer groups claim that the precautionary policies taken by the EU
toward GM food represent non-tariff trade barriers. One key piece of informa-
tion needed to help resolve this conflict is information about consumer demand
for GM foods in the US and EU. Results of this study tend to lend credence to
the argument made by EU regulatory agencies that Europeans are concerned
Valuation case studies 163
about the new technology. The welfare analysis conducted by Lusk et al. (2005)
using these auction bids implies, under a set of assumptions about prices and
segregation costs, that, on average, consumers in the three US locations have
benefited from the introduction of GM food even without a label, whereas con-
sumers in the two European locations would have, on average, suffered welfare
losses due to the advent of GM food had labeling not been implemented. Their
results show that a labeling policy would be welfare-reducing in the US, with the
opposite occurring in Europe. These findings correspond well with the policies
actually enacted in the US and Europe.
7.9.1 Experiment
We now examine the experimental design of Fox et al. (2002) that explores how
consumers respond to this type of controversy. They consider how favorable
and unfavorable information about food-irradiation affects willingness to pay
(WTP) for a reduction in the chance that trichinella is present in pork. The
favorable description emphasizes the safety and benefits of the process while
the unfavorable description stresses the potential risks. They elicit WTP values
for the upgrade from typical pork to irradiated pork and examined the changes
in elicited values following the introduction of new information about the tech-
nology. Values are elicited using a second price auction in which the binding
trial is chosen at random at the end of the experiment.
From a randomly selected sample of 200 households, eighty-seven primary
food shoppers were recruited to participate in a “consumer economics exper-
iment.” In return for participating, subjects were offered $40.00. No other
information about the experiment was provided to avoid participation bias
related to irradiation, or any other feature of the experiment. Participants were
assigned to ten groups ranging in size from six to twelve. There were three
treatments: the first examined the effect of positive information about irradia-
tion (POS), the second examined the effect of negative information (NEG), and
164 Experimental Auctions
the third examined the effect of providing both positive and negative informa-
tion simultaneously (BOTH). Two groups each were assigned to the POS and
NEG treatments with the remaining six groups assigned to the BOTH treat-
ment. All experiments were conducted in a food taste-testing lab at Iowa State
University.
The experiment contained several controls to ensure that participants under-
stood the auction procedures and to elicit ex ante and ex post attitudes. Initially,
subjects received the following descriptions of two pork sandwiches:
r Type I. This is a typical pork sandwich. The pork in this sandwich has a typical
chance of being contaminated with trichinella.
r Type II. The pork in this sandwich has been treated by irradiation to control
trichinella. Because of this treatment we can guarantee that this pork will not
cause trichinosis.
Subjects were informed that they had each been endowed with a typical pork
sandwich (Type I), and that an irradiated (Type II) pork sandwich would be
sold using a second price, sealed-bid auction in which the highest bidder wins
and pays the second highest bid. The auction had ten rounds (or trials) of
bidding, each with equal probability of being binding. In each trial, the monitor
publicly displayed the identification number of the highest bidder and three
bids: the second highest, average, and lowest bid. Subjects were informed that
in order to complete the experiment and earn their participation fee they would
have to consume, at the conclusion of the experiment, either: a) the typical
pork sandwich which they had been given; or b) the irradiated pork sandwich
which was available for purchase in the auction. Before the auction began, the
monitor provided each subject with the following “neutral” description of the
food irradiation process:
The U.S. Food and Drug Administration (FDA) has recently approved the use of
ionizing radiation to control Trichinella in pork products and Salmonella in poultry.
Based on its evaluation of several toxicity studies, the FDA concluded that irradi-
ation of food products at approved levels did not present a toxicological hazard to
consumers nor did it adversely effect the nutritional value of the product. Irradia-
tion of pork products at approved levels results in a 10,000 fold reduction in the
viability of Trichinella organisms present in the meat. The forms of ionizing energy
used in food processing include gamma rays, x-rays, and accelerated electrons. Ion-
izing energy works by breaking chemical bonds in organic molecules. When a suf-
ficient number of critical bonds are split in the bacteria and other pests in food, the
organisms are killed. The energy levels of the gamma rays, accelerated electrons,
and x-rays legally permitted for processing food do not induce measurable radioac-
tivity in food. This description is based on a review of the scientific literature on food
irradiation.
Favorable description
Food irradiation (also called ion pasteurization) is a process that destroys harm-
ful bacteria and pathogens by treating foods with ionizing radiation. Food
irradiation has been shown to be highly effective in destroying bacteria and
parasites responsible for food poisoning. Extensive research has proven that
irradiation is a safe and reliable process, and it has been approved by the Food
and Drug Administration, the American Medical Association, and the World
Health Organization. Each year as many as 9,000 people die in the US from
food-borne illness. Millions more suffer short term illness due to pathogens
such as salmonella, listeria and e.coli. By eliminating these pathogens from
food, irradiation can help to greatly reduce the number of food-borne illnesses.
This description is based on information supplied by the American Council on
Science and Health, a consumer education association.
Unfavorable description
Food irradiation is a process whereby food is exposed to radioactive materials,
and receives as much as 300,000 rads of radiation – the equivalent of 30 million
chest x-rays – in order to extend the shelf life of the food and kill insects
and bacteria. While it is unlikely that food products themselves will become
radioactive, irradiation results in the creation of chemicals called radiolytic
products, some of which are known carcinogens. Studies have also suggested
that irradiation may be linked to cancer and birth defects. Food irradiation can
kill most of the pathogenic bacteria present in food, but so can proper cooking.
Food irradiation was developed in the 1950s by the Atomic Energy Commission.
The objective was to seek potential uses for the byproducts of nuclear weapons
production. Today’s food irradiation industry is a private, for-profit business
enterprise with ties to the US nuclear weapons and nuclear power industries.
This description is based on information supplied by Food and Water, Inc., a
consumer advocacy group.
7.9.2 Results
Figure 7.3 shows the effect of new information about irradiation in the POS and
BOTH treatments on the bids for irradiated pork. Bids in trials 1 through 5 in
166 Experimental Auctions
Information treatment
Far safer 1 6 6 0 19 8
Somewhat safer 12 10 7 1 20 11
About as safe 4 2 6 10 9 15
Somewhat less safe 1 0 0 4 2 11
Far less safe 0 0 0 4 0 5
Total 18 18 19 19 50 50
each treatment are based on common information about irradiation – i.e., the
“neutral description” provided before any bids were submitted. To simplify the
figure we omitted the NEG treatment – those results were as expected with bids
for irradiated pork declining and with about 90% of participants bidding zero at
the end of the experiment. The favorable information also has the expected effect
and Figure 7.3 shows the increase in bidding following trial 5. But when the
same favorable and unfavorable descriptions are presented simultaneously, the
net impact is a significant reduction in bids for the irradiated product, with
the median bid falling to zero.
In addition to bidding for the irradiated pork, participants were also asked to
record their perception of its safety relative to the typical product. Table 7.15
summarizes those relative safety assessments both before and after the provision
of new information. The favorable information reinforces the perception that
Valuation case studies 167
7.9.3 Conclusions
The tendency to place more weight on negative information has potentially
important implications for public policy. For processes such as food irradia-
tion, genetic modification, and the use of growth hormones, even though the
scientific evidence is favorable, claims by opponents, even if they are inaccurate
and only suggest potential risks, will tend to reduce consumer demand. Because
the media tends to provide both sides of every story, the negative descriptions
created by advocacy groups tend to be widely available. In the US, even with
favorable scientific and regulatory opinion about the safety and wholesomeness
of irradiated foods, industry has been slow to introduce the process because it
fears the impact of inaccurate claims on consumer demand. The direct implica-
tion for public policy is that if new food technologies are generally regarded to
be in the public interest, then the anti-technology messages may reduce welfare.
168 Experimental Auctions
Constant −1.04
(0.766)
Prior attitude 0.08 0.487
1 = positive, (0.344)
0 = neutral;
−1 = negative
Gender: 0.33 0.126
Male = 1 (0.370)
Age: Categorical 5.82 −0.126*
2 = 20–24; (0.074)
4 = 30–34;
6 = 40–44;
8 = > 50; etc
Education: Categorical 5.53 0.166
3 = High school grad (0.101)
4 = Some trade school
5 = Some college
6 = BS/BA,
9 = Ph.D, etc
Positive treatment N = 18 2.599***
(0.472)
Negative treatment N = 19 −0.184
(0.469)
Correct predictions 74%
An even larger issue is the tradeoff between the free speech rights of opponents
of new technology compared with the potential loss in general welfare when
promising new technologies are halted. The results also indicate the enormous
responsibilities of those in the media who report on these controversies. If con-
sumer opinion is sensitive to negative information by consumer advocates, it is
likely they are sensitive to negative claims about food innovations made by gov-
ernments – even if scientists, perhaps the government’s own scientists, disagree.
In that context, recognition of consumer preferences as legitimate grounds for
protectionist trade policies may provide an incentive for governments to cast
doubt on new technologies when the goal is to protect domestic producers. That
scenario poses a legitimate threat to the expansion of world trade.
One open question is how such lab experiments on irradiated foods compare
to retail behavior in the wilds. People in the lab still know that they are being
monitored in a stylized setting, and the range of alternative purchases is more
Valuation case studies 169
limited than in a retail setting. In response, Shogren et al. (1999) compare the
similarity of lab valuation choices to retail store choices for irradiation. They
also compare choices made in hypothetical surveys. All subjects came from
the same small college town and made choices between typical chicken breasts
versus chicken breasts irradiated to reduce the risk of food-borne pathogens.
Their results from both the survey and experimental market suggested signif-
icantly higher levels of acceptability of irradiated chicken than in the retail trial
at an equal or discounted price for irradiation. Consumer choices were more
similar across market settings at a price premium for irradiation. They observed
that in a mail survey and a lab experiment, both of which included informa-
tion about irradiation, 80% of participants preferred irradiated to non-irradiated
chicken breasts when they were offered at the same price. When the irradiated
product was offered at a higher price, the survey and experimental results pre-
dicted market share in a subsequent retail trial remarkably well. About thirty
percent of survey respondents, experimental market participants, and shoppers
were willing to pay a ten percent premium for the irradiated chicken, and fif-
teen to twenty percent were willing to pay a 20% premium. Retail purchases
involved payment of real money for real products within an environment where
thousands of products competed for the consumer’s dollar. Any attempt to col-
lect consumer information in a retail setting was liable to interfere with the
realism of the market itself. The goal of the retail setting was to establish the
most realistic baseline possible against which one could judge the lab or survey.
Perfectly simulating a retail experience in the lab or a survey so as to control
every nuance is unattainable. Rather the goal should be to compare lab and sur-
vey choices relative to a real world baseline. The research program that emerges
for future work is to explore how comprehensive the lab or survey environments
must be to come closer to replicate an actual market outcome.
that 56.7% of the survey respondents were negatively inclined toward bST. This
negative reaction is not surprising, given that most people know little about
bST; furthermore, Smith and Warland (1992) believe the contexts in which the
surveys were conducted predisposed respondents somewhat negatively toward
bST.
The situation faced by an interviewee in a survey differs from that which
potential buyers will experience in retail settings. Many new products have
both positive and negative attributes (cheaper or leaner versus biotechnological)
that create ambiguity for consumers. Survey methods are hypothetical and the
relative lack of information and time that constrain the survey approach makes
it difficult to elicit accurate beliefs and can bias responses depending on how
questions are asked. For example, a study by Pitman-Moore, a company that
has developed a commercial pST, found that an average consumer would pay
more for leaner pST-treated pork than for fattier pork from an untreated animal.
In contrast, a study by Hoban and Burkhardt (1991) found that 45% of their
respondents were very concerned, and 37% were somewhat concerned, about
eating genetically engineered meat products.
perceptions of the type 1 and type 2 milk. After the tenth trial, the following
description of bST was provided:
Bovine somatotropin is a protein produced in the pituitary gland of a dairy cow that
regulates and stimulates milk production. Through advances in genetic engineering,
synthetic bST can now be manufactured using recombinant DNA technology. The bST
is injected into cows to increase milk yields. The frequency of these injections may range
from once a day to once every 14 to 28 days. Dairy cows treated with artificial bST have
produced from 10 to 25% more milk in experimental trials. They have also shown an
increase in feeding efficiency. The amount of bST in milk from treated cows has not been
shown to differ from that found naturally in milk. However, there is concern by some
people that too little research has been conducted to ensure the safety of milk and dairy
products from cows treated with bST. Bovine somatotropin is currently under regulatory
review and is expected to be approved soon by the Food and Drug Administration.
After twenty trials, a number was randomly drawn to determine which trial
would be binding and who would purchase the type 2 milk. Subjects were
required to drink the milk that they possessed at the end of the auction. The
experiments were carried out at five universities in Iowa, Arkansas, Mas-
sachusetts, and California. Fifteen undergraduate students from a range of
degree programs participated in an auction at each university. Because under-
graduate students comprise a specialized segment of the milk consuming pop-
ulation, results are not generalizable to the results to the general population.
Care was taken to replicate the experiments as closely as possible in a similar
environment at all five locations. All experiments were conducted by the same
investigator.
Dollars
2
1.5
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Trial
Figure 7.4 Average willingness to pay for “non-BST” milk in Iowa (),
Arkansas (+), Massachusetts (x), California (rural) (), California (urban)
(×)
With the exception of Iowa, the average bid fell between trials ten and eleven.
The drop in bids was greatest in urban California and was also large for Arkansas
and Massachusetts. This drop was most likely the result of the information
provided after trial ten. Participants were informed that milk from cows treated
with bST was very similar in composition to milk currently available at retail
outlets and that the FDA was expected to approve bST in the near future. The
effect of this information on the average bid of the urban California participants
exceeded $1.00. People from urban California had the least prior information
of any group of participants, and their behavior demonstrates the potential
sensitivity of consumers to the way in which information is presented. The
description of bST was a scientifically-balanced representation of the facts and
it served to eliminate almost completely the concerns among these uninformed
participants.
The behavior of the Massachusetts subjects was quite different. These par-
ticipants may have had more exposure to negative publicity about bST (e.g., the
Ben and Jerry’s campaign, which focused on the northeastern United States),
and had formed opinions they retained in spite of the information provided.
Neither the rural California nor Iowa experiments demonstrated any partici-
pant sensitivity to the information, which was to be expected as these were
Valuation case studies 173
California California
Bid($) urban rural Massachusetts Arkansas Iowa
0 6 9 6 7 12
0.01 to 0.25 4 0 2 1 0
0.26 to 0.50 2 4 0 1 0
0.51 to 0.75 1 2 0 1 0
0.76 to 1.00 1 0 0 1 0
≥ 1.00 1 0 7 4 3
7.10.3 Conclusions
These results suggest a relatively small group at each location would pay a large
premium to avoid drinking bST milk but that more than 60% would purchase
the product if it were available at the same price or at a slightly lower price
than milk from untreated animals. The results also support the view that how
information about these products is disseminated matters. In the absence of
174 Experimental Auctions
about – exploring the boundaries of behavior given control over bidding struc-
tures and given context over what this bidding means in broad social setting.
you may be unaware of some of these types of labels. To assist you in your
evaluation, we have provided the following steak descriptions.
Generic steak
The generic steak has no label. Beyond the fact that the steak has been US
Department of Agriculture (USDA) federally inspected, no other guarantees
are given regarding the quality of the meat. For your information, steaks similar
to this were selling at $8.09/lb last week at Dillon’s Supermarket. This would
be equivalent to $6.07 for a 12 oz. steak.
will ignore bids for the other two candy bars. Importantly, all candy bars
have an equally likely chance of being binding.
9) Once the binding round and candy bar have been determined, the winning
bidder will come forward and pay the 2nd highest bid amount and receive
winning candy bar. All other participants will leave with no candy bar.
Important notes
r You will only have the opportunity to win an auction for one candy bar.
Because we randomly draw a binding round and binding candy bar, you
cannot win more than one candy bar. That is, under no bidding scenario will
you take home more than one candy bar from this experiment.
r The winning bidder will actually pay money to obtain the winning candy bar.
This procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what it is worth to you to
obtain each of the four candy bars. Consider the following: if you bid more
than the candy bar is worth to you, you may end up having to buy a candy
bar for more than you really want to pay. Conversely, if you bid less than the
candy bar is really worth to you, you may end up not winning the auction
even though you could have bought a candy bar at a price you were actually
willing to pay. Thus, your best strategy is to bid exactly what each candy bar
is worth to you.
r It is acceptable to bid $0.00 for any candy bar in any round.
Example
Suppose there were five people participating in an auction just like the one you
are about to participate in. Suppose that these individuals participated in five
auction rounds, as you will, and that the 5th round was randomly selected to
be binding. Also, assume that the Nestlé Crunch bar was randomly selected to
be the binding candy bar. Now, suppose in round 5, participant #1 bid $0.00
for the Nestlé Crunch bar, participant #2 bid $0.10, participant #3 bid $0.25,
participant #4 bid $0.40, and participant #5 bid $0.50 for the Nestlé Crunch bar.
Who would win the auction? Participant #5 would win the auction because
they bid the highest amount. How much would participant #5 have to pay for
the Nestlé Crunch bar? Participant #5 would pay the 2nd highest bid amount,
which was $0.40. Thus, participant #5 would come to the front of the room, pay
$0.40 and obtain the Nestlé Crunch bar. Participants #1, #2, #3, and #4 would
pay nothing and would leave with no candy bar.
Note: these dollar amounts were used for illustrative purposes only and should
not in any way reflect what the candy bars may be worth to you.
Do you have any questions before we begin?
Please use the bid sheets, marked “candy bar auction.”
Valuation case studies 179
7.3.2 Instructions for steak auction Now that you have had the chance to
learn how the auction will work, we are interested in your preferences for
five different ribeye steaks. Each of you should have been given an information
sheet describing several different types of steaks. Please take a moment and read
with me through the information sheet. <<now read through the information
sheet>>
Now, we will give you the opportunity to participate in an auction to purchase
the type of steak you desire. Here in the front of the room, we have five steaks as
described in the information sheet. Again, other than differences in the labeled
characteristics, the steaks are the same size, weight, packaging, etc.
We will now conduct an auction for each of the steaks, where you will have
the opportunity to purchase one steak. In a moment, you will be asked to indicate
the most you are willing to pay (if anything) for each of the steaks by writing
bids on the enclosed bid sheets. The procedures for this auction are exactly the
same as the candy bar auction. To refresh your memory as to how the auction
works, I will go through the instructions again.
1) First, each of you has been given a bid sheet in your packet. On this sheet
you will, in a moment, write the most you are willing to pay to for each
of the following: a) the generic steak, b) the guaranteed tender steak,
c) the natural steak, d) the USDA choice steak, and e) the certified Angus
beef steak. Note: you will write five bids, one for each steak. Note: your
bids are private information and should not be shared with anyone.
2) After you’ve finished writing your bids, the monitor will go around the room
and collect the bid sheets.
3) In the front of the room, each of your bids will be ranked from highest to
lowest for each steak.
4) The person with the highest bid for each steak will win the auction and pay
the 2nd highest bid amount for the steak.
5) For each steak we will write the winning bidder number and the winning
price on the chalkboard for everyone to see.
6) After posting the price, we will re-conduct the auction for five additional
rounds.
7) At the completion of the 5th round, we will randomly draw a number 1
through 5 to determine the binding round. For example, if we randomly
draw the number 5, then we will ignore outcomes in all other rounds and
only focus on the winning bidder and price in round 5. Importantly, all
rounds have an equally likely chance of being binding.
8) After the binding round has been determined, we will randomly draw a
number 1 through 5 to determine which steak to actually auction (either
the generic steak, the guaranteed tender steak, the natural steak, the USDA
choice steak, or the certified Angus beef steak). For example, if we draw the
number 1, we will only focus on bids for the generic steak and will ignore
180 Experimental Auctions
bids for the other four steaks. Importantly, all steaks have an equally likely
chance of being binding.
9) Once the binding round and steak have been determined, the winning bidder
will come forward and pay the 2nd highest bid amount and obtain the
winning steak. All other participants will pay nothing and receive nothing.
Important notes
r You will only have the opportunity to win an auction for one steak. Because
we randomly draw a binding round and binding steak, you cannot win more
than one auction. That is, under no bidding scenario will you take home more
than one steak from this experiment.
r The winning bidder will actually pay money to obtain the winning steak. This
procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what each steak is worth to
you. Consider the following: if you bid more than the steak is worth to you,
you may end up having to buy a steak for more than you really want to pay.
Conversely, if you bid less than the steak is really worth to you, you may
end up not winning the auction even though you could have bought a steak
at a price you were actually willing to pay. Thus, your best strategy is to bid
exactly what the steak is worth to you.
r It is acceptable to bid $0.00 for any steak in any round.
r We realize that you may have errands to run after this experiment and may
not be able to take your steak immediately home. We offer two services to
accommodate your need. First, if you do not wish to take your steak home,
we will give you a coupon that can be redeemed at the K-State meat lab.
At some future date, you may come by and pick up the steak (your steak
will be of the quality obtained in this experiment). Or, if you wish, we can
give you a small styrofoam cooler to keep your steak cool until you get
home.
Do you have any questions before we begin?
Please use the bid sheets marked “steak auction.”
should have been given an information sheet describing several different types
of steaks. Please take a moment and read with me through the information sheet.
<<now read through the information sheet>>
Now, we will give you the opportunity to participate in an auction to purchase
the type of steak you desire. Here in the front of the room, we have five steaks as
described in the information sheet. Again, other than differences in the labeled
characteristics, the steaks are the same size, weight, packaging, etc.
We will now conduct an auction for each of the steaks, where you will have
the opportunity to purchase one steak. In a moment, you will be asked to indicate
the most you are willing to pay (if anything) for each of the steaks by writing
bids on the enclosed bid sheets. The procedures for this auction are exactly the
same as the candy bar auction. To refresh your memory as to how the auction
works, I will go through the instructions again.
1) First, each of you has been given a bid sheet in your packet. On this sheet
you will, in a moment, write the most you are willing to pay to for each of
the following: a) the generic steak, b) the guaranteed tender steak, c) the
natural steak, d) the USDA choice steak, and e) the certified Angus beef
steak. Note: you will write five bids, one for each steak. Note: your bids
are private information and should not be shared with anyone.
2) After you’ve finished writing your bids, the monitor will go around the
room and collect the bid sheets.
3) In the front of the room, each of your bids will be ranked from highest to
lowest for each steak.
4) Next, a random number will be drawn to determine how many participants
will win steaks. The random number will be somewhere between 1 and
the total number of participants. Call this random number N.
5) The N − 1 highest bidders will win the auction and all winning bidders
will pay the Nth highest bid amount for the exchange. For example, sup-
pose there were 10 participants that submitted bids and the number 4 was
randomly drawn by the monitor (i.e., N = 4). In this case, the 3 (N − 1)
highest bidders will win the auction and each will pay the 4th highest bid
amount for the winning steak.
6) For each steak we will write the winning participants’ bidder numbers, the
random number (N), and the winning price on the chalkboard for everyone
to see.
7) After posting the prices and winning bidder numbers, we will re-conduct
the auction for five additional rounds.
8) At the completion of the 5th round, we will randomly draw a number 1
through 5 to determine the binding round. For example, if we randomly
draw the number 5, then we will ignore outcomes in all other rounds and
only focus on the winning bidders and price in round 5. Importantly, all
rounds have an equally likely chance of being binding.
182 Experimental Auctions
9) After the binding round has been determined, we will randomly draw a
number 1 through 5 to determine which steak to actually auction (either the
generic steak, the guaranteed tender steak, the natural steak, the USDA
choice steak, or the certified Angus Beef steak). For example, if we draw
the number 1, we will only focus on bids for the generic steak and will
ignore bids for the other four steaks. Importantly, all steaks have an equally
likely chance of being binding.
10) Once the binding round and steak have been determined, the winning
bidders will come forward and pay the Nth highest bid amount for the
winning steak. All other participants will pay nothing and will not receive
a steak.
Important notes
r You will only have the opportunity to win an auction for one steak. Because
we randomly draw a binding round and binding steak, you cannot win more
than one auction. That is, under no bidding scenario will you take home more
than one steak from this experiment.
r The winning bidders will actually pay money for the winning steak. This
procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what each steak is worth to
you. Consider the following: if you bid more than the steak is worth to you,
you may end up having to buy a steak for more than you really want to pay.
Conversely, if you bid less than the steak is really worth to you, you may
end up not winning the auction even though you could have bought a steak
at a price you were actually willing to pay. Thus, your best strategy is to bid
exactly what the steak is worth to you.
r It is acceptable to bid $0.00 for any steak in any round.
r We realize that you may have errands to run after this experiment and may
not be able to take your steak immediately home. We offer two services to
accommodate your need. First, if you do not wish to take your steak home,
we will give you a coupon that can be redeemed at the K-State meat lab.
At some future date, you may come by and pick up the steak (your steak
will be of the quality obtained in this experiment). Or, if you wish, we can
give you a small styrofoam cooler to keep your steak cool until you get
home.
Do you have any questions before we begin?
Please use the bid sheets marked “steak auction.”
here as they very closely mirror the instructions for the steak auctions – see, for
example, the detail in section A7.3 for the second price auction.
Important notes
r You will only have the opportunity to win an auction for one steak. Because
we randomly draw a binding round and binding steak, you cannot win more
than one auction. That is, under no bidding scenario will you take home more
than one steak from this experiment.
r The winning bidder will actually pay for the winning steak. This procedure
is not hypothetical.
r In this auction, the best strategy is stay in the auction until the point at which
the price is exactly equal to what the steak is worth to you.
r It is acceptable to immediately raise the sheet with your participant ID number
when the auction begins. In this case, you are willing to pay $0.00 for the
steak.
r We realize that you may have errands to run after this experiment and may
not be able to take your steak immediately home. We offer two services to
accommodate your need. First, if you do not wish to take your steak home,
we will give you a coupon that can be redeemed at the K-State meat lab. At
some future date, you may come by and pick up the steak (your steak will be
of the quality obtained in this experiment). Or, if you wish, we can give you
a small styrofoam cooler to keep your steak cool until you get home.
Do you have any questions before we begin?
Important notes
r You will only have the opportunity to win an auction for one steak. Because
we randomly draw a binding round and binding steak, you cannot win more
than one auction. That is, under no bidding scenario will you take home more
than one steak from this experiment.
r The winning bidder will actually pay money for the winning steak. This
procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what each steak is worth to
you. Consider the following: if you bid more than the steak is worth to you,
186 Experimental Auctions
you may end up having to buy a steak for more than you really want to pay.
Conversely, if you bid less than the steak is really worth to you, you may
end up not winning the auction even though you could have bought a steak
at a price you were actually willing to pay. Thus, your best strategy is to bid
exactly what the steak is worth to you.
r It is acceptable to bid $0.00 for any steak in any round.
r We realize that you may have errands to run after this experiment and may
not be able to take your steak immediately home. We offer two services to
accommodate your need. First, if you do not wish to take your steak home,
we will give you a coupon that can be redeemed at the K-State meat lab. At
some future date, you may come by and pick up the steak (your steak will be
of the quality obtained in this experiment). Or, if you wish, we can give you
a small styrofoam cooler to keep your steak cool until you get home.
Do you have any questions before we begin?
Please use the bid sheets marked “steak auction.”
Auction procedures
1) First, each of you has been given a bid sheet in your packet. On this sheet
you will, in a moment, write the least amount of money you are willing to
sell your candy bar for. Your bids are private information and should not be
shared with anyone.
2) After you’ve finished writing your bids, the monitor will go around the room
and collect the bid sheets.
3) In the front of the room, bids for the candy bar will be ranked from lowest
to highest.
4) The person with the lowest bid for selling their candy bar will win the auction
and be paid the 2nd lowest bid selling price.
5) We will write the winning bidder number and the winning price on the
chalkboard for everyone to see.
Valuation case studies 187
6) The winning bidder will come forward and be paid the 2nd lowest bid
amount in exchange for his/her candy bar. All other participants will be able
to keep their candy bar, but will not be given any other chance to sell their
candy bar for money.
Important notes
r The winning bidder will actually receive real money and forfeit their candy
bar. This procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what your candy bar is worth
to you. Consider the following: if you bid less than the candy bar is worth to
you, you may end up selling your candy bar for less than you really value it.
Conversely, if you bid more than the candy bar is really worth to you, you
may end up not winning the auction even though you could have sold the
candy bar at a price you were willing to receive. Thus, your best strategy is
to bid exactly what each candy bar is worth to you.
r It is acceptable to bid $0.00 if the candy bar has no worth to you.
Example
Suppose there were five people participating in an auction just like the one you
are about to participate in. Now, suppose in the bidding round, participant #1
bid $0.00, participant #2 bid $0.10, participant #3 bid $0.25, participant #4 bid
$0.40, and participant #5 bid $0.50 for the sale of their candy bar.
Who would win the auction? Participant #1 would win the auction because
they bid the lowest amount. How much would participant #1 be paid for the
Snickers bar? Participant #1 would be paid the 2nd lowest bid amount, which
was $0.10. Thus, participant #1 would come to the front of the room, receive
$0.10 in exchange for their candy bar. Participants #2, #3, #4, and #5 would
receive nothing and would leave with their candy bar.
Note: these dollar amounts were used for illustrative purposes only and should
not in any way reflect what the candy bars may be worth to you. Do you have
any questions before we begin? Please use the bid sheets, marked “candy bar
auction.”
To refresh your memory as to how the auction works, I will go through the
instructions again.
Auction procedures
1) You will fill out your financial records inventory sheet which will be col-
lected by the monitor.
2) You will then fill out your bid sheets by submitting your total selling price
for your financial information listed on the inventory sheet. You will also
fill out on the bid sheet how much each financial document counted towards
your final valuation.
3) In the front of the room, each of your bids will be ranked from lowest to
highest.
4) The person with the lowest bid for their financial records will win the auction
and will be paid the 2nd lowest bid amount.
5) We will write the winning bidder number and the winning price on the
chalkboard for everyone to see.
6) The winning bidder will come forward and identify himself/herself.
Important notes
r This experiment will take place in several different locations across Indiana.
The lowest bidder from all of the experiments combined will receive money
in exchange for their records. The winner at each auction site will need to
identify themselves and give their contact information in case they are the
overall winner.
r The overall winner will actually receive money for their financial records.
This procedure is not hypothetical. We will expect the winner to give us all
of their original records and all duplicative copies of the records they sold.
We will keep the records in our possession for five years and then destroy
them. Only <<project director>> will have access to the records that are
sold. The winner of the auction will no longer have access to your financial
records once they are in our possession. The winner will be allowed to have
access to his/her tax and employment records if he/she is audited.
r In this auction, the best strategy is to bid exactly what your financial records
are worth to you. Consider the following: if you bid less than your records
are worth to you, you may end up losing your financial records. Conversely,
if you bid more than your records are really worth to you, you may end up
not winning the auction even though you could have sold your records at a
price you were willing to take. Thus, your best strategy is to bid exactly what
your financial records are worth to you.
Please use the sheets marked “Financial Records Inventory” and “Bid sheet
for Financial Records Auction.”
190 Experimental Auctions
Auction procedures
1) First, each of you has been given a bid sheet in your packet. On this sheet
you will, in a moment, write the least amount of money you must be paid
to exchange your Large Snickers bar for the Small Snickers bar. This is the
amount of money I must pay you so that you’d be willing to give up your
Large Snickers bar and take the Small Snickers bar instead. Note: Your bids
are private information and should not be shared with anyone.
2) After you’ve finished writing your bids, the monitor will go around the room
and collect the bid sheets.
3) In the front of the room, bids will be ranked from lowest to highest.
4) The four lowest bids for the Small Snickers bar will win the auction. The
individuals that submitted the four lowest bids will be paid the 5th lowest
bid amount to exchange their Large Snickers bar for the Small Snickers.
5) The numbers of the winning bidders and the winning price (the 5th lowest
bid) will be written on the chalkboard for everyone to see.
6) After posting the price, I will re-conduct the auction for two additional
rounds.
7) At the completion of the 3rd round, I will randomly draw a number 1 through
3 to determine the binding round. For example, if we randomly draw the
192 Experimental Auctions
number 3, then we will ignore outcomes in all other rounds and only focus
on the winning bidder and price in round 3. Importantly, all rounds have an
equally likely chance of being the winning round.
8) Once the binding round has been determined, the winning bidders will come
forward and will be paid the 5th lowest bid amount and exchange their Large
Snickers for the Small Snickers. All other participants will be paid nothing
and will keep their Large Snickers.
Important notes
r You will only have the opportunity to win an auction for one candy bar.
Because we randomly draw a binding round, you cannot win more than one
candy bar. You will either leave with one Large Snickers if you are a high
bidder or one Small Snickers and some amount of money if you are a low
bidder.
r I will actually pay money to the winning bidders to exchange their Large
Snickers for the Small Snickers bar. This procedure is not hypothetical.
r In this auction, the best strategy is to bid exactly what it is worth to you to
exchange your Large Snickers for the Small Snickers. Consider the following:
if you bid more than it is worth to you to exchange your Large Snickers bar
for a Small Snickers bar, you may end up not winning the auction even
though you could have exchanged your Large Snickers at a price you were
actually willing to be paid. Conversely, if you bid less than it is worth to
you to exchange your Large Snickers for the Small Snickers, you may end
up having to exchange your Large Snickers at a price lower than you really
wanted to. Thus, your best strategy is to bid exactly what it is worth to you
to exchange your Large Snickers bar for the Small Snickers bar.
r It is acceptable to bid $0.00 for the Small Snickers in any round. This would
mean that you are willing to give up your Large Snickers for the Small
Snickers without any compensation from me.
r Importantly, we are interested in what it is worth to you to exchange your
Large Snickers for the Small Snickers. We are not interested in your total value
of the Small Snickers, only the amount of the exchange. In other words, what
is the premium you would place on your Large Snickers versus the Small
Snickers?
Auction example
Suppose there were ten people participating in an auction just like the one you
are about to participate in. Suppose that these individuals participated in three
auction rounds, as you will, and that the 2nd round was randomly selected to be
binding. Now, suppose in round 2, participant #1 bid $0.00 to exchange their
Large Snickers bar for the Small Snickers, participant #2 bid $0.03, participant
#3 bid $0.06, participant #4 bid $0.09, and participant #5 bid $0.12, participant
Valuation case studies 193
Participant
Number #1 #2 #3 #4 #5 #6 #7 #8 #9 #10
Bid Amount $0.00 $0.03 $0.06 $0.09 $0.12 $0.15 $0.18 $0.21 $0.24 $0.27
Who would win the auction? Participants #1, #2, #3, and #4 would win the
auction because they were the four lowest bidders. How much would I pay
participants #1, #2, #3, and #4 to give up their Large Snickers bar and take the
Small Snickers? I would pay them the 5th lowest bid amount, which was $0.12.
Thus, participants #1, #2, #3, and #4 would come to the front of the room, be paid
$0.12 and exchange their Large Snickers bar for a Small Snickers. Participants
#5, #6, #7, #8, #9 and #10 would be paid nothing and would leave with their
free Large Snickers bar.
* Note: these dollar amounts were used for illustrative purposes only and
should not in any way reflect what the candy bars may be worth to you*
Do you have any questions before we begin?
Please use the bid sheets marked “candy bar auction.”
the candy bar auction, with one exception: we request that all participants eat
their cookies at the end of the auction. To refresh your memory as to how the
auction works, I will go through the instructions again.
Auction procedures
1. First, each of you has been given a bid sheet in your packet. On this sheet
you will, in a moment, write the least amount of money you must be paid
so that you would be willing to exchange your cookie for the cookie that
contains genetically modified ingredients. This is the least amount of money
I must pay you to take the cookie with genetically modified ingredients
instead of the cookie you now have, which contains NO genetically modified
ingredients. Note: your bids are private information and should not be shared
with anyone.
2. After you’ve finished writing your bids, the monitor will go around the room
and collect the bid sheets.
3. In the front of the room, each of your bids will be ranked from lowest to
highest.
4. The four lowest bids will win the auction. The individuals with the four
lowest bids will be paid the 5th lowest bid amount for the exchange.
5. We will write the winning bidder numbers and the winning price on the
chalkboard for everyone to see.
6. After posting the price, we will re-conduct the auction for nine additional
rounds.
7. At the completion of the 10th round, we will randomly draw a number 1
through 10 to determine the binding round. For example, if we randomly
draw the number 5, then we will ignore outcomes in all other rounds and
only focus on the winning bidder and price in round 5. Importantly, all
rounds have an equally likely chance of being binding.
8. Once the binding round has been determined, the winning bidders will come
forward and be paid the 5th lowest bid amount and exchange their cookie for
the cookie made with genetically modified ingredients. All other participants
will keep their cookie made without genetically engineered ingredients.
Important notes
r You will only have the opportunity to win an auction for one cookie. Because
we randomly draw a binding round, you cannot win more than one cookie
from this auction.
r The winning bidder will actually be paid money to exchange their cookie with
no genetically modified ingredients for the cookie with genetically modified
ingredients. This procedure is not hypothetical.
r We will also expect every participant to eat their cookie at the conclusion of
the auction.
Valuation case studies 195
r In this auction, the best strategy is to bid exactly what it is worth to you to
exchange your cookie for the cookie with genetically modified ingredients.
r It is acceptable to bid $0.00 in any round. This would mean that you are
willing to accept the cookie made with genetically modified ingredients for
no compensation.
r Importantly, we are interested in what it is worth to you to exchange your
cookie for the cookie made with genetically modified ingredients.
Do you have any questions before we begin?
8 Auction design: case studies*
8.1 Introduction
Results from experimental auctions frequently generate as many questions as
answers about bidding behavior and auction design. Creating an auction that
balances experimental control and real-world context can raise fundamental
questions about how experience with a good and the market affects bidding,
whether preferences are fixed or fungible with different market interaction,
how incentive structures can affect bidding behavior, and how hypothetical
payments affect bidding behavior. This chapter reports on some of our own
work to explore questions of auction design; queries triggered by attempts
to value new food products and other basic goods. We focus on eight case
studies related to auction design: preference learning, auction institution and the
willingness to pay/willingness to accept gap, second price auction tournaments,
fixed or fungible preference, gift exchange, calibration of real and hypothetical
bidding, and bidding behavior in consequential auctions.
∗ This chapter draws on material worked on over the past decade with colleagues and students
including Dermot Hayes, Sean Fox, Greg Parkhurst, Chris McIntosh, Bob Wilhelm, Sara Gun-
nersson, Todd Cherry, Cannon Koo, Michael Margolis, John List, and Lucine Tadevosyan.
196
Auction design: case studies 197
3
2
Bidit = αi + β St + CX it + εit (8.1)
1 1
198 Experimental Auctions
where Bidit denotes subject i’s bid in trial t, α i represents a fixed/random bid-
der effect, which controls for individual-specific time-invariant effects such as
subjects’ private values and subjective risk perceptions for the good; St are
dichotomous variables that represent each of the four data gathering sessions
(excluding Session 4 as the baseline reference category); and Xit are other fac-
tors that may affect bidding behavior. In Xit , we include lagged price to control
for any posted price effects, and a dichotomous variable that indicates whether
the participant was a winner of the item in a previous session (winner = 1;
otherwise 0).
Table 8.1 contains our panel data estimation results. The results suggest that
bidding behavior differed across sessions, but only for unfamiliar goods. Coeffi-
cient estimates in the candy and mango regressions show that bidding behavior
did not significantly change across the four sessions for these two goods. These
results imply bidders were not enticed to bid differently on familiar goods
when they are confronted with the novel lab environment. Empirical estimates
in the pork regression differ. Subjects tend to decrease their bids slowly over
the four sessions. For example, a coefficient estimate of 0.051 for the session 1
dummy variable indicates that controlling for price, whether the subject was
a previous winner, and subject-specific effects, participants bid about a nickel
higher in the first session compared to the fourth. Given that the mean bid
in all four sessions was approximately nine cents, this decrease is dramatic.
Auction design: case studies 199
Combining empirical results from the pork regression with estimates from
regressions of the familiar goods suggests that people tend to bid in an aggres-
sive manner in early sessions, but only for unfamiliar goods. For familiar goods
that the bidder may have had past experience, there appears to be little evidence
of a lab novelty effect. We reject the novelty of the experiment experience
hypothesis.
The results also show that conditional bids from winners significantly
decrease after experience with the good. Estimates related to the coefficient
of winner in each auction imply that winners tend to reduce their bids sig-
nificantly in the next session. Conditioning on subject-specific effects, lagged
price, and session number, our estimates imply that winning subjects decrease
their bid by about 3.3, 14.1, and 10 cents in the sessions after they purchased
the candy, mango, and pork. Because the sessions were held several days apart
for goods that are probably consumed in or shortly after the experiment, the bid
reduction is probably not a result of movement down the demand curve. Instead,
results support the notion that people are trying to learn their preference for the
unfamiliar good.
The results of this study indicate that preference learning about unfamiliar
goods explains the “high” bids in experimental auctions, not the novelty of the
experimental experience. Behavior is consistent with the view that a person’s
bid includes an information value that reflects his desire to learn more about
unfamiliar goods, i.e., preference learning. Of course, this information value is
a real value that might be expected to be present in the marketplace. Depend-
ing on the purpose of an auction, the information value may be an important
economic component to measure (say when estimating market share of a novel
good). Other studies may be more interested in longer-run valuations when all
consumers are fully informed; if so, the design of the study presented in this
section provides a way to disentangle the preference learning value from the
underlying value for the good. Additional research on the relationship between
individual measures of familiarity with a good and bidding behavior would help
to further resolve this issue.
Evidence has accumulated over the past two decades suggesting a significant
gap exists between WTP and WTA. Consider the conflicting results from two
visible papers. First, Kahneman et al. (1990) report evidence to support their
idea of the endowment effect as why WTP diverges from WTA. The idea behind
the psychological argument for the endowment effect is that people treat gains
and losses asymmetrically: the fear of a loss weighs more heavily than the enjoy-
ment from an equal gain. The effect exists if people offer to sell a commonly
available good in their possession at a substantially higher amount than they
will pay for the identical good not in their possession. Their results make a case
for the existence of a fundamental endowment effect as WTA exceeded WTP
in all treatments over all iterations. People’s preferences seemed to depend on
initial endowments of resources, a violation of rational choice theory.
In contrast, Shogren et al. (1994) designed an experimental auction to test
the proposition that with positive income elasticity and repeated market partic-
ipation, WTP and WTA will converge for a market good with close substitutes
(e.g., candy bars and mugs) but will not converge for a good with imperfect
market substitute (a sandwich with health risk). They also considered whether
transaction costs could explain the WTP-WTA gap by providing the market
goods (e.g., candy bars and mugs) immediately outside the lab door that could
be purchased by the subjects once the experiment ended. Their results showed
that WTP and WTA values converged after several repeated trials for candy bars
and mugs (goods with many available substitutes), but that the values continued
to diverge for the reduced health risks from safer food.
The question is why different results were obtained in these two studies.
One major distinction between Kahneman et al. (1990) and Shogren et al.
(1994) is their choice of auction mechanism. Kahneman et al. (1990) used the
BDM mechanism to elicit preferences whereas Shogren et al. (1994) used a
second price auction. The choice of auction should not matter in theory since
a bidder’s weakly dominant strategy is to state their true WTP or WTA in both
mechanisms. Evidence exists, however, which reveals differences in bids across
incentive-compatible auctions. Rutström (1998), for instance, detected WTP
bid differences elicited in the BDM and Vickrey style auctions. Differences in
auction mechanism might be one cause of the differences in results between
these two studies.
Did the choice of auction mechanism generate the different findings on the
WTP-WTA gap? Too many differences exist across the two original experimen-
tal designs to say for sure. In response, Shogren et al. (2001b) addressed this
issue by designing an experiment in which the auction mechanism was varied
as a treatment variable. If the endowment effect accounts for observed behavior,
the effect should be observable and persistent for any mechanism used to elicit
WTP and WTA, provided the mechanism is incentive compatible. Shogren et al.
(2001b) tested this thesis by evaluating the impact of three auction mechanisms
Auction design: case studies 201
in the measurement of WTP and WTA for goods with close substitutes: the
BDM mechanism with exogenous price feedback, second price auction with
endogenous market-clearing price feedback, and a random nth price auction
with endogenous market-clearing price feedback.
Table 8.2 summarizes the design parameters in Kahneman et al. (1990),
Shogren et al. (1994), and the new Shogren et al. (2001b) study. The key
experimental parameters in the new design are: (i) auctioned goods: a brand-
name candy bar in Stage 1 and an Iowa State University coffee mug in Stage 2;
(ii) initial monetary endowment: $15 paid up-front; (iii) number of trials: ten
trials per experiment, in which wealth effects were controlled by randomly
selecting one of the ten trials to be the binding trial; (iv) retail price information
on the commodities: none was provided; (v) subject participation: voluntary
participants from the student population at Iowa State University; (vi) number
of subjects per session: eight to ten subjects in the second price auctions and
twenty for the BDM mechanism; and (vii) the auction mechanism: either the
BDM, second price, or random nth price.
For the auction mechanisms, the BDM mechanism in Kahneman et al.’s fifth
experimental treatment was used. The BDM mechanism worked as follows.
Step 1: Twenty subjects were randomly divided into two subgroups of ten
buyers and ten sellers. Step 2: Monitors gave each seller a candy bar in Stage 1
(a coffee mug in Stage 2), which he or she could take home or sell at the auction.
Step 3: Each buyer had the option to buy the commodity in the auction. Step 4:
Each buyer and seller were asked to determine independently and privately their
maximum WTP or minimum WTA by marking an “X” on a recording sheet
listing a price schedule such as:
I will buy (sell) I will not buy (sell)
If the price is $ 0.40
If the price is $ 0.30
If the price is $ 0.20
Step 5: After collecting all recording sheets from buyers and sellers, one price
from the sheet was selected randomly. Step 6: If a buyer was willing to pay
at least the exogenous random price, he or she bought the commodity. If the
seller was willing to accept less than or equal to the random price, he or she
sold the commodity. Step 7: The random price along with the number of buyers
and sellers willing to buy and sell at the random price were recorded on the
blackboard as public information. Step 8: After the tenth trial in each stage,
one of the ten trials was randomly selected as the binding trial that determined
take-home pay.
The second price auction was the same as in Shogren et al. (1994). In the
WTP treatment, bidders were asked to record, privately and independently, the
Table 8.2. Summary of experimental design parameters
Design parameter Original Kahneman et al. (1990) Original Shogren et al. (1994) New experiments Random nth price auction
Auctioned goods Tokens, pens, & mugs Candy bar, sandwich, & mugs Candy bar & mugs Candy bar & mugs
Initial monetary endowment None $3: candy bar $15 $15
$15: sandwich or mug
Number of trials Varied between 3 and 7 5: candy bar 10: candy bar 10: candy bar
20: sandwich 10: mugs 10: mugs
10: mugs
Retail price information Provided for some treatments None provided None provided None provided
Subject participation In-class Voluntary Voluntary Voluntary
Number of subjects per Varied between 30 & 44 12 to 15 8–10: second 10: random nth price
session price auction auction
20: BDM
Simon Fraser U. Iowa State U. Iowa State U. U. Central FL
Auction institution Becker-DeGroot-Marschak Second price auction (SPA) Both the BDM Random nth price auction
Mechanism (BDM) and the second
price auction
Auction design: case studies 203
maximum he or she was willing to pay for the candy bar (Stage 1) or coffee mug
(Stage 2) on a recording sheet in ten repeated trials. In each trial, the monitors
collected the recording cards, and then posted the identification number of the
highest bidder and the market-clearing price (the second highest bid) on the
blackboard as public information. One of the ten trials was randomly selected
as binding at the conclusion of the session. In the WTA treatments, monitors
gave each subject a candy bar (Stage 1) or a coffee mug (Stage 2). Subjects
wrote their minimum WTA to sell the commodity on their recording sheet in
each of ten trials and for both stages. Again monitors posted the identification
number of the lowest bidder and the market-clearing price (the second lowest
bid) after each trial. After the tenth trial in each stage the monitors randomly
selected one of the ten trials as the binding trial. The auctions were conducted
in three sessions on the same day at Iowa State University. There were 78
participants. Different subjects participated in each session. Each session was
completed within two hours.
Tables 8.3 and 8.4 present the summary statistics of the experimental results
on the BDM and second price auction treatments. In the BDM treatment, the
average selling price exceeded the average buying price for all ten trials in both
the candy bar and coffee mug stages. The same is true of the median bids. The
average and median WTA/WTP ratios remain relatively constant throughout
the ten trials, ranging from 1.5 to 1.8 for the candy bar and 1.4 to 1.6 for the
coffee mug. The null hypothesis that the mean WTP and WTA are equal can
be rejected at the p < 0.01 level (p = 0.05 and p = 0.10) for candy bars (coffee
mugs) using a small sample t-test and a Mann-Whitney U-test.
The bidding behavior in the second-price auction followed a different pat-
tern. As illustrated in Table 8.4, bidding behavior in trial 1 does not contradict
Kahneman et al.’s (1990) concept of an endowment effect. The average and
median WTA bid was significantly greater than the average or median WTP
bid in trial 1. Unlike the BDM treatment, however, WTA offers decreased and
WTP bids increased after the initial trial in each second price auction. The aver-
age and median WTA/WTP ratios decline throughout the ten trials, decreasing
from 1.4 in trial one to 0.88 in trial nine for the candy bar (the ratio in trial ten
was 4, including an extreme outlier bid of $50).1 The average WTA-WTP ratio
decreased from 2.5 in trial one to 1.05 in trial ten for the coffee mug. In trials
two to ten for the candy bar, we cannot reject the null hypothesis that the mean
WTP and WTA are equal at the 10% level using a t-test and at the 5% level using
a Mann-Whitney U-test. For the coffee mug, we cannot reject the null in trials
three to seven and nine to ten at the 5% level using the t-test, and we cannot
reject the null in trials two to ten at the 1% level using the Mann-Whitney U-test.
The value disparity faded away in the second-price auction with feedback on
the endogenous market-clearing price: average WTP and WTA value measures
1 The second price auction was clearly not demand revealing in trial ten for this subject.
Table 8.3. Summary statistics of the Becker-DeGroot-Marschak Mechanism
Trial
Candy WTP Mean $0.58 $0.50 $0.54 $0.54 $0.53 $0.53 $0.47 $0.48 $0.51 $0.51
N = 20 Median 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50
Variance 0.121 0.093 0.071 0.070 0.080 0.084 0.097 0.092 0.075 0.079
WTA Mean 0.92 0.90 0.86 0.82 0.83 0.78 0.80 0.79 0.78 0.78
N = 20 Median 0.90 1.00 0.95 0.85 0.80 0.80 0.80 0.80 0.80 0.80
Variance 0.067 0.086 0.072 0.105 0.073 0.067 0.087 0.092 0.091 0.114
Ratio of mean WTA/WTP 1.59 1.80 1.59 1.52 1.57 1.47 1.70 1.65 1.53 1.53
t-test of means −3.5a −4.2a −3.8a −2.9a −3.4a −2.9a −3.4a −3.2a −2.9a −2.7a
Mug WTP Mean $2.40 $2.40 $2.25 $2.28 $2.25 $2.25 $2.23 $2.23 $2.10 $2.18
N = 20 Median 1.75 1.75 1.75 1.75 1.75 1.75 1.75 1.75 1.25 1.50
Variance 4.200 4.700 4.540 4.144 3.961 3.961 3.802 3.802 3.085 3.928
WTA Mean 3.68 3.55 3.50 3.40 3.40 3.35 3.40 3.43 3.35 3.13
N = 20 Median 3.25 3.50 3.25 3.00 3.00 3.00 3.00 3.00 3.00 3.00
Variance 5.955 5.287 5.553 5.726 5.674 5.134 5.463 5.349 5.108 5.628
Ratio of mean WTA/WTP 1.53 1.46 1.56 1.49 1.51 1.49 1.52 1.54 1.60 1.44
t-test of means −1.8b −1.6c −1.8d −1.6e −1.6e −1.6e −1.7b −1.8d −2.0d −1.4e
Trial
Candy WTP Mean $0.56 $0.62 $0.63 $0.70 $0.68 $0.72 $0.69 $0.71 $0.73 $0.75
N = 18 Median 0.53 0.63 0.62 0.75 0.72 0.80 0.70 0.78 0.86 0.90
Variance 0.090 0.086 0.092 0.102 0.098 0.174 0.103 0.114 0.118 0.123
WTA Mean 0.78 0.84 0.65 0.62 0.89 0.67 0.51 0.64 0.64 3.00
N = 20 Median 0.75 0.60 0.50 0.45 0.43 0.37 0.29 0.29 0.29 0.26
Variance 0.169 1.033 0.376 0.363 4.631 1.199 0.530 1.267 1.278 123.57
Ratio of mean WTA/WTP 1.39 1.35 1.03 0.89 1.31 0.93 0.74 0.90 0.88 4.00
t-test of means −1.9b −0.9 −0.1 0.5 −0.4 0.2 1.0 0.3 0.3 −0.9
Mug WTP Mean $2.22 $2.73 $2.90 $3.19 $3.38 $3.34 $3.33 $3.58 $3.42 $3.28
N = 18 Median 1.75 3.00 3.25 3.40 3.63 3.53 3.74 3.80 3.87 3.73
Variance 2.142 1.407 1.573 1.805 1.855 2.821 2.726 2.276 2.918 2.687
WTA Mean 5.52 4.56 3.77 3.58 3.52 3.16 3.08 2.45 3.33 3.44
N = 20 Median 4.88 5.00 4.48 3.75 4.01 3.18 3.25 1.13 3.50 4.00
Variance 15.912 13.397 6.971 6.775 5.749 6.364 6.448 5.498 7.073 8.304
Ratio of mean WTA/WTP 2.49 1.67 1.30 1.12 1.04 0.95 0.92 0.68 0.97 1.05
t-test of means −3.4a −2.1c −1.3d −0.6 −0.2 0.3 0.4 1.8c 0.1 −0.2
are not statistically different. If the idea that the effect should persist across
auctions and trials is correct, our results suggest no fundamental endowment
effect exists.
Although compelling, one could challenge the results with notions of affili-
ated bidding behavior and a top dog effect. First, concern has been voiced that
most bidders walk into a valuation auction cold, and thus the common uncer-
tainty about the value of the good might create “affiliated” values or beliefs
(see Milgrom and Weber, 1982). Affiliation exists if high bids induce low bid-
ders to increase their bids. The posted bid affects behavior because prices send
information to bidders about commonly perceived, but unknown, characteris-
tics of the product. List and Shogren (1999) reject this affiliation hypothesis for
familiar goods in second price auctions with feedback. Second, some observers
have suggested that the second price auction creates an environment that is too
competitive to elicit people’s true WTA and WTP. People might be submitting
bids to win for winning’s sake. They might bid WTP up and WTA down to leave
the experiment as the “top dog” among their peers. While this is a theoretical
possibility, experimental evidence at this time provides little support for this
conjecture (see Shogren et al., 2001b). The evidence suggests that the top dog
effect is not a universal phenomenon in valuation experiments, and the case
remains unarticulated by its supporters as to why it should exist for one set of
goods over another.
But these retorts do not explain the observations of Knetsch et al. (2001) who
ran second price and ninth price auctions with repeated trials for a familiar good.
They found large gaps in WTP and WTA values for the ninth price auction, but
not for the second price auction. They concluded that the “endowment effect
remains robust over repeated trials, and that contrary to common understanding
the Vickrey auction may not be demand revealing.” They make the reasonable
case that the uniform auction is not necessarily the preferred mechanism if it
fails to engage off-the-margin bidders, i.e., bidders whose value is far below or
above the market-clearing price. They argue something is fundamentally flawed
about the mechanism, that is, the endowment effect is not the “problem” the
mechanism is the “problem.”
These off-the margin bidders in the second and ninth price auctions might
have less incentive to make sincere bids if they believe they have no chance to
win or to lose the auction and profit. The second price auction might not engage
low value bidders, while the ninth price auction might not engage high value
bidders. Ninth price auction bidders realize the support of bids is significantly
below their value and therefore can afford to submit false bids without being
punished by the market. Evidence using induced values does not contradict this
conjecture: bidders off-the-margin of the market-clearing price frequently do
not bid sincerely (e.g., Miller and Plott, 1985). Off-margin low value bidders
in the second price auctions are bounded from above by the high value bidders;
Auction design: case studies 207
off-margin high value bidders in the ninth price auction are unbounded from
above. This suggests that the off-margin bidding behavior has greater chance
for more variability in the ninth price auction.
To address this question of unengaged bidders, Shogren et al. (2001b) tested
the random nth price auction. The auction attempts to engage all bidders by
combining the best parts of the BDM and second price mechanisms: a random
but endogenously determined market-clearing price. Randomness to keep all
bidders in the auction and reduce their incentive to chase a particular market-
clearing price; endogenous to guarantee that the bidders themselves determine
the market-clear price.
The random nth price auction works as follows: (i) each bidder submits a
bid (offer); (ii) each bid (offer) is rank-ordered from lowest to highest; (iii)
the monitor selects a random number, the n in the nth price auction, uniformly
distributed between two and N (N bidders); and (iv) in the WTP case, the
monitor sells one unit of the good to each of the (n − 1) highest bidders
at the nth price; in the WTA case, the monitor buys one unit each from the
(n − 1) lowest bidders and pays the nth lowest bid. Lab evidence suggests the
auction can work: bidders revealed sincere bids, and off-the-margin bidders
were more engaged relative to the standard second price mechanism (Shogren
et al., 2001a).
Shogren et al. (2001b) constructed the random nth price treatment to match
the two other treatments except for the obvious switch of the auction mechanism.
Table 8.2 summarizes the random nth price experiments, which were run in four
sessions on the same day at the University of Central Florida with a total of forty
subjects. Table 8.5 summarizes the experimental results. Bidding behavior was
similar to that observed in the second price auction. The initial WTA-WTP gap
disappears with experience. The average and median WTA and WTP bids are
significantly different through trial three for the candy bar and trial four for the
coffee mug, which does not contradict Kahneman et al.’s (1990) concept of an
endowment effect.
The results then show the WTP and WTA values converged in the last seven
trials for candy bars and six trials for coffee mugs. For the candy bar, the median
WTA/WTP ratio dropped from nearly two to one from trial one to trial ten. We
cannot reject the null hypothesis that the mean WTP and WTA are equal at
even the 10% confidence level using a t-test or a nonparametric Mann-Whitney
U-test in trials four to ten. For the coffee mug, the average WTA/WTP ratio
decreased to 0.78 in trial ten from 2.42 in trial one. We cannot reject the null in
trials five to ten at the 10% level using the t-test, and we cannot reject the null in
trials two to ten at the 5% level using the Mann-Whitney U-test. Like behavior
in the second price auction, little value disparity is observed in the random nth
price auction after minimal market experience: average WTP and WTA value
measures are not statistically different in the later trials.
Table 8.5. Summary statistics of the random nth price auction
Trial
Candy WTP Mean $0.69 $0.70 $0.68 $0.74 $0.68 $0.88 $0.91 $0.93 $1.01 $0.99
N = 20 Median 0.53 0.58 0.50 0.50 0.61 0.75 0.77 0.88 1.00 1.00
Variance 0.30 0.26 0.33 0.45 0.31 0.45 0.48 0.51 0.47 0.52
WTA Mean 1.29 1.33 1.26 1.07 0.93 1.05 1.06 1.05 0.88 0.88
N = 20 Median 1.00 1.13 1.12 1.00 1.00 0.99 0.88 0.87 0.99 1.00
Variance 0.67 0.64 0.56 0.45 0.21 0.98 1.14 1.20 0.22 0.20
Ratio of mean WTA/WTP 1.87 1.90 1.85 1.44 1.36 1.20 1.15 1.13 0.87 0.89
t-test of means −2.7a −2.9a −2.8a −1.5 −1.5 −0.67 −0.48 −0.41 0.70 0.58
Mug WTP Mean $2.15 $2.21 $2.32 $2.21 $2.27 $2.76 $2.99 $2.77 $2.99 $3.83
N = 20 Median 1.00 1.00 1.23 1.55 2.00 2.25 2.00 3.00 2.50 3.25
Variance 6.07 6.06 6.22 4.55 2.88 3.72 7.38 2.76 3.10 4.96
WTA Mean 5.21 4.62 4.19 4.29 3.63 3.35 3.37 2.63 2.98 3.02
N = 20 Median 5.03 4.50 3.75 3.53 3.00 2.50 2.03 2.00 2.35 2.00
Variance 12.69 11.83 10.98 12.94 11.72 5.97 6.09 4.64 5.70 5.80
Ratio of mean WTA/WTP 2.42 2.08 1.81 1.93 1.60 1.21 1.12 0.95 0.99 0.78
t-test of means −3.1a −2.5a −2.0d −2.2b −1.6 −0.84 −0.46 0.23 0.01 1.10
In conclusion, Shogren et al. (2001b) find that while initial bidding behavior
does not contradict the endowment effect concept, the effect can be eliminated
with repetitions of a second price or random nth price auction. These results
provide little support that the endowment effect exists insofar as the endow-
ment effect is perceived as an effect that should persist across auction mech-
anisms and across trials. These findings suggest the auction mechanism could
have accounted for the conflicting observations in Kahneman et al. (1990)
and Shogren et al. (1994). These results raise the question of whether the
endowment effect is a fundamental part of choice or simply an artifact of a
weak exchange environment (see also Plott and Zeiler, 2005). The weaker
the exchange institution, the weaker the socialization of rational behavior and
the stronger the potential hold of asocial anomalies on choice. Determining
exactly why three theoretically incentive compatible mechanisms are not all
demand-revealing suggest more research is needed to develop a theory of
auction-dependent bidding behavior and more experimental work to test out its
predictions.
For the standard auction, the expected payoff was $21.85, with maximum and
minimum earnings of $25.50 and $18.50, including the flat $5 participation fee
for each subject. For the tournament treatment, the expected payoff was $30.00,
with a maximum and minimum of $120 and $5.2
2 Some evidence exists that subjects react differently to differences in absolute payoffs provided
differences are substantial enough. This could suggest the expected payoff difference for the
tournament treatment could induce bidders to pay more attention to bidding just because of
Auction design: case studies 211
For each round, induced values were randomly drawn from a uniform distri-
bution on [$0.00, $20.00] in ten-cent increments. Each auction had ten bidders;
two sessions were run for each treatment for a total of forty bidders. Each ses-
sion lasted about forty minutes. Table 8.6 summarizes bidding behavior for all
rounds across both treatments. The unconditional results suggest the tourna-
ment induced more sincere bidding than the standard design. Mean bids were
more accurate, and the variation was smaller with the tournament. The mean bid
in the standard treatment exceeded the mean induced value (i.e., resale value)
by $6.28 (standard deviation = $63.51); the mean bid in the tournament was
only $0.96 larger than induced value (standard deviation = $4.14).
Using conditional panel regression analysis, three key results emerge. First,
bidders bid more sincerely in the second price auction tournament than in the
standard second price auction. Sincere bidding in the tournament is tested with
a two-way random effects model:
Bidit = φIN it + αi + ϕt + εit , (8.2)
where Bidit denotes subject i’s bid in trial t; INit denotes subject i’s induced
value in trial t; α i represents subject-specific characteristics; ϕ t represents
the absolute level effect. It remains an open question whether the less than twofold difference
we have in our treatments is substantial enough to make a difference. Interestingly, one could
have equated expected payoffs by increasing the show-up fee to $12.15 in the standard auction
treatment; but it is unclear whether this would have made the incentives for sincere bidding better
or worse given an even flatter payoff in the standard auction. In addition, the tournament payoffs
were stated in dollar terms, whereas the standard auction was presented as points to be converted
to dollars.
212 Experimental Auctions
Tournament auction
sample Standard auction sample Pooled sample
The slope of the regression equation is twice as steep as theory predicts, and the
intercept term below zero. The null hypothesis (that α = 0 and ϕ = 1) is rejected
at the 6% significance level. The rejection of the null hypothesis is the first
indication that the tournament with the non-linear payoff may be structurally
different than the standard auction with linear payoffs. Indeed, using a Chow
test, we find the estimated models differ significantly across the two treatments:
bidders were less likely to bid sincerely in the standard auction (F1,796 = 5.87,
p-value = 0.02).
Second, the tournament auction design is more effective at engaging on-
margin and off-margin bidders than the standard auction design. The effec-
tiveness of the tournament relative to the standard auction to engage both on-
and off-margin bidders is examined by separating the induced values into three
groups: off-margin low values [$0, $5]; off-margin mid values ($5, $10]; and on-
margin values ($10, $20]. Extending (8.2), the following equation is estimated
to disentangle the treatment effects on sincere bidding:
where the dependent variable, Bidit , is subject i’s bid in trial t, α i is a subject-
specific fixed/random effect that accounts for systematic differences in bidding
patterns and controls for ordering of the experiment types; INit is subject i’s
induced private value in trial t; NLit = 1 for non-linear pay (tournament) treat-
ment, 0 otherwise; and midit = 1 for bidder i in trial t if his value was in the
range ($5, $10] in trial t, 0 otherwise; and lowit = 1 for bidder i in trial t if
his value was in the range [$0, $5] in trial t, 0 otherwise. The remaining vari-
ables INit *NLit , midit *INit , midit *NLit, midit *INit *NLit , lowit *INit , lowit *NLit ,
and lowit *INit *NLit , are interaction terms and allow slope and intercept changes
across payment method and bidder type (i.e., off- or on-the-margin bidders).
Table 8.8 presents the regression results. Based on the Hausman test, we
estimate a random effects model, which includes the NL dummy variable (col-
umn 3). Evaluating the estimated regression equations for high induced values,
indicates the slope on the standard treatment is roughly four times that of the
tournament treatment, 3.90 versus 0.92 (the tournament slope is close to demand
revelation, i.e., unity). The intercept of the regression line is 2.7 for the tourna-
ment and −32.35 for the standard auction. A Wald test is used to test the joint
hypothesis of demand revelation, intercept = 0, slope = 1. We reject the null
hypothesis for the standard treatment (χ22 = 13.32, p-value < 0.01), but fail to
reject the null hypothesis for the tournament (χ22 = 0.10, p-value = 0.95). For
214 Experimental Auctions
Random effects
Variable Random effects Fixed effects W/treatment variable
Standard and tournament equal χ52 = 5.29 χ52 = 6.57 χ62 = 7.83
(0.62) (0.75) (0.75)
Table 8.9. Efficiency in the tournament and standard second price auction
Tournament A 10 3 10 8
Tournament B 12 4 7 11
n = 40
Total 22 7 17 19
Percentage 55% 17.5% 42.5% 47.5%
Standard Auction A 6 7 9 4
Standard Auction B 11 4 5 5
n = 40
Total 17 11 14 9
Percentage 42.5% 27.5% 35% 22.5%
auction tournament; and 42.5% in the standard auction. Second, in the auction
tournament, the high-value bidder either won or set the price in 97.5% of all
rounds; in the standard auction, this occurred in only 77.5% of rounds. Third,
in the tournament, the bidder that should have set the market-clearing price did
so 47.5% of the time, compared to just 22.5% in the standard auction. For strict
efficiency – the auction is strictly efficient when the highest value bidder wins the
auction and pays the bid of the player with the second highest induced value –
we observe the auction tournament accurately predicts the winner and price
setter in 45% of all rounds, whereas this only occurs half as frequently (22.5%)
in the standard auction. Focusing on the last ten rounds, the predicted winner and
the price setter was observed 75% in the tournament, and 25% in the standard
auction.
If one also considers efficiency as measured by the percentage of maximum
surplus captured (max surplus = highest induced value – second highest induced
value), we find that efficiency is highest for the last ten rounds in the tourna-
ment, capturing 66% of the potential surplus. By this measure of efficiency,
the standard treatment was negative in rounds one to ten and eleven to twenty,
which suggests the gains captured by a winning highest induced value bidder
were more than offset by the losses of winning bidders with low induced values.
Why did the second-price tournament generate more rational behavior rela-
tive to the standard auction? On the surface, a tournament setting should create
a more complex strategic setting than a standard auction. The fear as stated by
Camerer and Hogarth (1999, p. 36) is “in theory, tournament incentives should
induce more status-seeking and risk-taking and hence, does not lead subjects
to incentive-compatibly maximize expected profit (which is why economists
usually eschew them). Whether tournaments actually do have those unintended
effects have not been carefully investigated.” We did not see such unintended
effects: we found the opposite.
These results are consistent with three behavioral explanations. First, the
tournament payoffs were framed such that winning points as clearly indicated
in the dollar payoffs (not just points to be converted into dollars) made points
appear more valuable in the tournament. Earning points in the tournament
could be viewed as more valuable so the typical mistake of overbidding and
occasionally suffering losses had more impact in the tournament. Four players
in the tournament stood to earn more than the average, where the winner and
runner up would earn substantially more. That the contingent expected payoffs
were slightly higher in the tournament could have been enough to trigger more
sincere bidding. Second, we purposefully created our auction tournament as
a repeated game of incomplete information with almost common knowledge.
Players did not know their relative position in the tournament regardless of
the round. The set of induced values differed randomly in each trial, only the
winning bidder knew he or she had won and the market price and cumulative
Auction design: case studies 217
profits were private. While such a game can lead to multiple equilibria, the
logic of a unique focal equilibrium arises based on risk dominance (e.g., select
the safest strategy: the one that maximizes one’s payoff independent of what
the other players do) and iterated deletion of dominated strategies (Kajii and
Morris, 1997). As points accumulate across rounds, the dominant strategy in the
last round of the second price tournament is to follow Vickrey’s initial strategy:
bid one’s induced value. Overbidding is dominated because it increases the risk
of paying too much and losing accumulated points; underbidding is dominated
because it increases the risk of missing out on profitable opportunities. Since
a bidder only learns his relative ranking at the game’s end, if he or she works
backwards eliminating dominated strategies, this leads to a weakly dominant
strategy of bidding one’s induced value in each round.
A third behavioral explanation rests in Heiner’s (1983) theory on the ori-
gins of predictable behavior. This theory suggests the more uncertainty, the
more predictable a person’s behavior. He notes “‘behavioral rules’ . . . arise
because of uncertainty in distinguishing preferred from less-preferred behav-
ior. Such uncertainty requires behavior to be governed by mechanisms that
restrict the flexibility to choose potential actions, or which produce a selec-
tive alertness to information that might prompt particular actions to be chosen.
These mechanisms simplify behavior to less-complex patterns, which are easier
for an observer to recognize and predict” (p. 561). Here the only information
each player had in a bidding round was his or her induced value (and own
accumulated points). Without information on relative ranking or the current
“tournament leader”, a player had no legitimate chance to strategize against
an opponent; rather he or she had to rely on a behavioral rule. Our results do
not contradict that the behavioral rule is to bid one’s induced value. Given the
limited ability of people to eliminate simple dominated strategies (see Camerer,
2003), our results do not contradict the Heiner story. One implication is that
rather than trying to explain irrational bidding ex post with new restrictions
on preferences, one can construct ex ante a second price auction environment
where sincere bidding is a good predictor of behavior.
matters for theory and public policy because if preferences are “transient arti-
facts” contingent on context, so are the welfare measures used in cost-benefit
analyses to rationalize or reject regulations to protect health and safety.
Gunnarsson et al. (2003) investigate the stability of preferences of people
who fall prey to the classic anomaly of preference reversals (e.g., a person
prefers lottery A to lottery B, but then puts greater monetary value on B than
A). In Chapter 9, we discuss how preference reversals have been documented for
isolated individuals in numerous lab experiments run by both economists and
psychologists (Grether and Plott, 1979; Tversky et al. 1990). One explanation
of preference reversals argues that the decision maker constructs preferences on
the spot instead of ranking the options. The strategies when constructing pref-
erences include (Slovic, 1991, p. 500), “anchoring and adjustment, relying on
prominent dimension, eliminating common elements, discarding nonessential
differences, adding new attributes into the problem frame in order to bolster one
alternative, or otherwise restructuring the decision problem to create dominance
and thus reduce conflict and indecision.”
Gunnarsson et al. (2003) used experimental data to estimate an empirical
model of preferences for risk and skewness, the love of the long shot, in market-
like and non-market settings. Their results show that preferences remained
stable even as arbitrage removed preference reversals. People stopped reversing
preferences with arbitrage not because their preferences were fungible, but
because they initially overpriced the risky long shot.
We use data from a lab experiment with 123 subjects (Cherry et al., 2003).
After entering the lab, participants signed a consent form acknowledging their
voluntary participation while agreeing to abide by the instructions. Written pro-
tocols ensured uniformity across sessions, and all subjects were inexperienced
with preference reversal experiments. The protocol included randomly seating
the subjects as they entered the room, disallowing any communication whatso-
ever among subjects, reading the experimental instructions aloud as the subjects
followed along, administering a test of comprehension, addressing any ques-
tions or concerns raised by the subjects, and conducting the market sessions.
Treatment 1: In this baseline treatment, subjects faced two independent set-
tings that create conditions likely to induce people to reverse their preferences.
In each setting, subjects were faced with two monetary lotteries: a p-bet lottery
and a $-bet lottery. A p-bet is a relatively safe lottery with a high probability
of winning a smaller reward; a $-bet is a relatively risky lottery with a low
probability of winning a larger reward. Subjects were asked which lottery they
preferred, and how much they valued each of the two lotteries. Preferences
and values were binding in both settings, e.g., subjects were sold lotteries for
their indicated value using a BDM mechanism. But inconsistent preferences
and values were not arbitraged. The process was repeated with fifteen different
lotteries.
Auction design: case studies 219
where g denotes all lottery games except G which is the highest top prize game,
PGj is the probability of winning the highest top prize game G in option j,
Uij (XGj ) is player i’s utility from winning the top prize XGj in game G in option
n
j, and Pgj ·Uij (X gj ) captures all other lottery games g offered in option j except
g=1
the highest top prize game in game G in option j. This term is the probability
of winning the top prize in any game g multiplied by player i’s utility from
winning the top prize, Xgj , in any game g summed over all n of g games. The
two terms in equation 8.4 reflect the expected utility for player i for all lottery
games available in option j, Uij .
Normalize utility Uij (XGj ) = 1 and impose the preferences restriction that the
odds between the gambles are selected to make the lottery players indifferent
220 Experimental Auctions
= β0 + β1 X gj + β2 X gj
2
+ β3 X gj
3
(8.8)
Pgj
where β 1 measures the bettor preferences over the mean of returns, β 2 measures
bettor risk aversion (β 2 > 0 risk loving; β 2 < 0 risk aversion; β 2 = 0 risk
neutrality), β 3 measures the bettor’s preference for skewness (β 3 > 0 favorable
preference for skewness; β 3 < 0 unfavorable preference for skewness; β 3 = 0
indifferent preference for skewness). If β 1 > 0, β 2 < 0, and β 3 > 0, then lottery
players are risk averse, and choose to play those lotteries with greater skewness
of returns. Because the experiment generates panel data, results expand on
previous work by modifying the empirical specification in two ways. First, the
independent variables are interacted with period indicator variables to estimate
any change in subject preference and risk aversion over time. Second, time
invariant subject attributes are controlled with a random effects specification.
Table 8.10 reports the results and shows the existence of the preference rever-
sal phenomenon in this laboratory setting. Results also show how the introduc-
tion of arbitrage significantly decreases the rate of reversals (i.e., increases
the rate of rational choices). Prior to arbitrage (round 6), reversal rates across
treatments were not significantly different at any standard level. After four
rounds of arbitrage, reversal rates were significantly lower in the arbitrage treat-
ments relative to no-arbitrage baseline (p-values < 0.02). In the later rounds,
the discipline from arbitrage was highly significant in generating more rational
choices.
Now we turn to the issue of subjects’ love of skewness and whether pref-
erences remained stable when people adjusted behavior to act more rationally.
Tables 8.11–8.13 report panel estimates of equation 8.8 across treatments. For
each treatment, the market and non-market setting, a separate model is esti-
mated. To estimate how preferences and risk aversion change within a setting
Auction design: case studies 221
Market Non-market
, † ,and ‡ indicate significance at the 10, 5 and 1 percent levels with the null being the rever-
sal rate in the arbitrage treatment (2 or 3) is equal to the rate in the non-arbitrage baseline
(treatment 1).
Note: arbitrage was introduced in round 6.
over time, we interact the measures (β 1 , β 2 , & β 3 ) with time period indicator
variables. Recall β 1 measures the preferences over the mean of returns, β 2 mea-
sures the preference for risk, and β 3 measures the preference for skewness. We
expect subjects are risk averse and prefer lotteries with greater expected payoffs
and higher skewness (β 1 > 0, β 2 < 0, and β 3 > 0). Our null hypothesis is that
risk aversion and skewness preferences remain stable even as people correct
preference reversal behavior in the face of a new context: market-like arbitrage.
Results in the market-like and non-market settings across all three treat-
ments confirm our expectations. In each model, estimated coefficients for the
early periods were significantly different than zero and carried signs consistent
with our expectations. Subjects preferred lotteries with higher expected returns
(β 1 > 0), were risk averse (β 2 < 0), and preferred lotteries with higher skewness
(β 3 > 0). One might imagine a link between the love for skewness and prefer-
ence reversals, but the results from the later periods contradict this notion. In the
later periods, estimated coefficients remained significantly different than zero
indicating that subjects were still risk averse and preferred higher expected
returns and skewness. But as the reversal rates declined in the market and
222 Experimental Auctions
Market Non-market
β1 β2 β3 β1 β2 β3
R̄ 2 0.552 0.542
N 615 615
Market Non-market
β1 β2 β3 β1 β2 β3
R̄ 2 0.540 0.540
615 615
Market Non-market
β1 β2 β3 β1 β2 β3
R̄ 2 0.532 0.532
N 615 615
Material value A D
Sentimental value
Total value x B C
columns reflect the study treatments: the hypothetical open-ended survey, the
hypothetical demand revealing auction, and the actual demand revealing auc-
tion. Cell A represents the hypothetical material value elicited in the Waldfogel
(1993; 1996) and Solnick and Hemenway (1996) surveys. Cell D reflects the
real material value we use to estimate the welfare loss/gain from gift giving.
Moving to cell D from cell A is complicated because the real auction occurs
in cell C. In cell C, we elicit total value because the credibility of the auction
would be stretched beyond reasonable limits to force a subject to forego his or
her sentimental value in a real sale. Since sentimental values are embedded in
the real offers, we use a three-stage experimental design to move from cell A
to cell D via cells B and C.
In Stage 1 (February 4), a monitor administered Waldfogel’s original survey
(cell A) to a group of 46 undergraduate students at the University of Central
Florida. After reading the experimental instructions and answering all relevant
questions, the monitor asked each subject to complete the holiday gift giving
survey. Information gathered from the survey included a description of all the
gifts the subject received at Christmas, the relationship the subject had with
each gift giver, and the estimated price the giver paid for each gift. Each subject
also stated the hypothetical material values to sell his or her gifts. A subject
who received ten gifts submitted ten values. Finally, each subject revealed the
additional sentimental value for each gift.
In Stage 2 (February 6), the monitor introduced the hypothetical Vickrey-
style uniform price, sealed-bid auction. Since most subjects received several
gifts, we used the random nth price auction to provide incentive for a subject
to take care in valuing each of his or her gifts. A second price auction was not
used because such a mechanism might cause a subject to focus on selling only
his or her lowest-valued gift. The goal was to increase the odds that gifts could
actually be sold that would otherwise be off-the-margin, and thereby give a
person more incentive to submit sincere offers for these gifts.
By using the random nth price auction, each subject could potentially sell
more than one gift at the uniform price. While truth-telling can be a (non-
unique) Nash equilibrium in a multiple-unit, uniform price auction, in general
this need not be the case (see Forsythe and Isaac, 1982). A multi-good, uniform
Auction design: case studies 227
price auction does not necessarily inherit the same demand revealing properties
as a second price or nth price auction in which subjects can only sell one gift
(Vickrey, 1962). While subjects have incentive to bid sincerely on the first gift
sold, their incentive to inflate the value of subsequent gifts increases because
winning sellers affect the market price with positive probability (see Ausubel
and Cramton, 1996). The subject can increase the expected selling price on the
first gift by inflating offers for additional gifts. Experimental evidence suggests
that nth price auctions do a reasonable job of empirical demand revelation on
aggregate (see Franciosi et al., 1993). Offers off the margin are less likely to
reveal demand truthfully, however, than those on the margin, as suggested by
theory. Alternative designs to remove this incentive include a variation of the
auction in which a bidder who sells K gifts receives the amount of the kth
lowest rejected offer other than his or her own offer for the kth object won, or
a descending-offer design that replicates the Vickrey auction (see Ausubel and
Cramton (1996) for details).
The auction works as follows: (i) for each gift received, a subject states his or
her total value to sell the gift; (ii) all gifts, gi , from eachsubject are then pooled
together to create the set of total available gifts, G = iM gi , where M is the
total number of subjects; (iii) all gifts, G, are then rank-ordered from lowest
to highest total value; (iv) the monitor selects a random number uniformly
distributed between 2 and 21 (21 was the most gifts received by a subject); and
(v) the monitor then purchases the (n − 1) lowest total value gifts overall and
pays the nth lowest total value for each gift. For example, suppose G = 500 gifts
overall and #6 was chosen as the random nth price, then only the five lowest
valuation gifts overall would be purchased at the sixth lowest offer.
After the instructions were read, the monitor ran a candy bar pre-auction to
give the subjects some experience with the random nth price auction. Next, we
ran the hypothetical auction in which each subject offered a selling price, or her
total value for each gift (cell B). Finally, the monitor asked the subject to split
this hypothetical total value into its sentimental and material value components
for each gift.
In Stage 3 (February 11), we ran the actual random nth price auction for
the gifts. After the monitor read the instructions and stated that this auction
was real, each subject submitted her real total value for each gift (cell C). The
subject again divided total value into its sentimental and material value (cell
D) components for each gift. The random price was #4, and the monitor made
arrangements with the sellers to purchase the three lowest total value gifts each
priced at the fourth lowest total value. Three distinct subjects sold one gift each
at a uniform price of $2.00. Although some attrition occurred over the eight
days of the three stage experiment, this is not a substantive issue since students
were unaware that a real auction would occur after the first two stages. Also,
gifts that were used before the experiment (e.g., candy, gift certificates, cash)
228 Experimental Auctions
a Waldfogel’s (1996) results are in brackets under his (1993) estimates; NA denotes not available.
b Solnick and Hemenway’s trimmed sample results are in brackets under full sample estimates.
c Standard error of the mean in parentheses beneath the yield estimates.
d “Mean yield across subjects” is the mean MV/TC for each subject averaged over all subjects.
e “Mean yield across gifts” is the MV/TC for each gift averaged over all gifts.
were trimmed from the analysis for obvious reasons. Our final sample consists
of 244 gifts across thirty-six recipients.
According to Waldfogel’s hypothetical survey given in Stage 1, gifts generate
a welfare loss: average yields across subjects and gifts were 99% and 98% (see
Table 8.15). Percentage yield across subjects is computed as the mean ratio of
material values to the total costs for each subject averaged over all subjects;
Auction design: case studies 229
percentage yield across gifts is the same ratio for each gift averaged over all gifts.
Although our results do not exactly replicate Waldfogel’s in that they do not
differ significantly from 100% at conventional levels, the point estimates do not
refute his finding that a deadweight loss may exist from gift-giving at Christmas.
These findings suggest that sample selection may not be the pivotal issue in this
debate. Subjects from the University of Central Florida respond as those at Yale.
At issue is the gap between hypothetical and actual behavior. In the actual
random nth price auction, we estimate a welfare gain: average yields were 121
and 135% (Table 8.15). Our results support Solnick and Hemenway’s intuition
that gift-giving does not universally destroy value, but our results suggest their
estimated gains of 214 to 10,100% (partial and full sample) are an upper bound.
Plausible value creation can occur with gift-giving.
The welfare gains are driven more by the actual auction than framing the
offers as demand revealing. This finding is illustrated in Columns 1–3 of
Table 8.16, in which summary statistics for the revealed total cost and mate-
rial value are shown for each of the treatments. Although mean yields differ
across the two hypothetical instruments, using a Wilcoxon signed-rank test for
matched pairs we cannot reject the hypothesis that the revealed values (material
value, sentimental value, or total value) in the hypothetical survey were derived
from the same parental population as the values from the hypothetical auction
(see Table 8.16, Wilcoxon test, W = 3281 (z = −0.46); W = 1120 (z = −1.09);
W = 4752 (z = −0.66)), we can reject the hypotheses that revealed values in
the actual auction are derived from the same parental populations as the two
hypothetical treatments (Table 8.16). The change to real economic commit-
ments from a hypothetical scenario seems to be the catalyst for the welfare
gain. If we calibrated Waldfogel’s results by increasing his material values with
the estimated mean real-hypothetical gap of 26.5–27.4% (from Table 8.16),
his deadweight loss becomes a surplus. Similar implications are drawn if the
calibration factor is taken from Table 8.15.
These results indicate that hypothetical surveys show a welfare loss from
Christmas gift giving, whereas the actual auction produced a plausible welfare
gain, as real selling prices were at least 27% higher than hypothetical prices. The
subjects reacted differently when the auction was real, a common phenomenon
noted in much of the non-market valuation literature. The relationship between
in-kind transfers and value is not immune to the gap between intentions and
actions, suggesting lessons learned in experimental auctions have implications
for broader economic questions of general public interest.
required. The majority of research shows the average person exaggerates his
actual willingness to pay when asked a hypothetical question (e.g., Bohm,
1972, Bishop and Heberlein, 1979, Dickie et al., 1987, Shogren, 1990, Seip
and Strand, 1992, Neill et al., 1994). This hypothetical bias has motivated
research into whether one can calibrate hypothetical and actual values using
experimental markets (e.g., Fox et al., 1998).
We now consider the experiment auctions in List and Shogren (1998), who
explored calibration by comparing bidding behavior in a hypothetical and actual
second price auction for baseball cards: deliverable objects with an intangible
quality. Baseball cards have many favorable characteristics for a calibration
exercise including familiarity, the ability to deliver, and an abstract quality
beyond the normal market good. List and Shogren (1998) ran three treatments
at a sports card show in Denver, CO: one card, one card among ten, and one
Auction design: case studies 231
card bid on by sports card dealers presumed more experienced with the market
than the general population (the Dealer treatment).
For the one-card and Dealer treatments, the auctioned good was a Cal Ripken
Jr. 1982 Topps Traded PSA graded nine rookie baseball card. All treatments
displayed the same Cal Ripken Jr. card to ensure comparability of bids. An
independent agency, Professional Sports Authenticators (PSA), graded the Cal
Ripken Jr. card to avoid complications of participants not understanding the
grade (i.e., substance and quality) of the card. We selected the Cal Ripken Jr.
card to reduce the valuation issues encountered when subjects do not understand
the substance of the good they are asked to value (Cummings et al., 1986). For
the ten-good treatment, we used the Cal Ripken Jr. card plus nine other sports
cards (or sets of cards) that could act as potential substitutes or complements,
e.g., Billy Ripken’s 1989 rookie PSA graded card, three Troy Aikman cards, one
complete set (without Aikman), and one Michael Irvin card. Bidders submitted
a separate bid for each of the ten cards.
The experimental auction for the one-good and ten-good treatments used a
four-step experimental design: (i) inspection of the good(s), (ii) hypothetical
bid(s), (iii) actual bid(s), and (iv) debriefing. In Step 1, monitor A approached
a person entering the show and asked if he or she would like to participate in
a hypothetical auction that would take about ten minutes. If the person agreed,
the monitor briefly explained that we were hypothetically auctioning off the
baseball card(s) displayed on the table. The participant could pick up and visu-
ally examine each card. All cards were sealed with the PSA grade clearly
marked on each cardholder. The monitor worked one-on-one with the partic-
ipant and no time limit was imposed on his or her inspection of the card(s).
We did not give the participants any financial incentives or gifts to participate,
thus we avoid any claims of the results being influenced by “found money”
effects.
In Step 2, monitor A gave the participant an instruction sheet that consisted of
two parts: (i) a short socio-economic survey (e.g., age, education, years trading),
and (ii) a bidding sheet. The participant was asked to submit a hypothetical
bid stating the maximum that he or she was willing to pay for the card(s).
The instructions for the bidding sheet stated that the hypothetical exchange
mechanism was a sealed-bid second price auction. The bidding sheet reported:
A sealed bid second price auction will be used to determine the winner of this item.
Thus, if your bid of $X for this item is the highest bid and the next highest bid is $X-5,
you win this item but will only pay $X-5. Under this bidding mechanism it is best for
you to bid your true value for this item because overbidding may cause you to pay too
much and underbidding decreases your odds of winning the item.
Note: You will not be required to pay this amount and all bids are hypothetical. Also,
the winner will not receive this card.
232 Experimental Auctions
After the participant filled out the survey and hypothetical bidding sheet pri-
vately, he or she folded the bidding sheet and placed it in an opaque box. The
monitor told the participant that his or her bid would not be opened until after
the show and that all bids would be destroyed when our research project was
complete. Monitor A then asked the participant to go over to monitor B at a
second table fifteen feet away for a follow-up auction.
In Step 3, monitor B told the participant that he or she now has the chance to
actually bid on the card(s) that he or she had just examined in Step 2. Monitor
B gave the participant a second bidding sheet for the actual auction. Again the
sealed bid second price auction was used as the exchange mechanism. After
the monitor answered all questions about the auction, the participant placed his
or her sealed bid into a second opaque box. To guarantee that we did not get a
second hypothetical bid monitor B asked each participant to acknowledge their
actual bid with a signature and valid telephone number where they could be
contacted. Care was taken to avoid contamination of the results by any ordering
effects (e.g., sealed boxes, monitors not handling or observing bids).
Finally, in Step 4 monitor B debriefed the participant. The monitor explained
that the participant would be contacted within three days after the show if he or
she was the highest bidder. Monitor B also explained that if the participant won
the auction, he or she would receive the card(s) after he or she had sent a check
or money order for the amount of the second highest bid. After any remaining
questions were answered, monitor B asked the participant not to discuss the
auction with anyone else until after the show, and then thanked him or her for
participating in the project. Within three days, the winner of each auction was
notified by phone, and when the monitors received the checks, they mailed out
the cards.
The one-good and ten-good treatments took approximately twelve hours to
complete (9 am to 9 pm). On the top of each hour the auction treatment was
switched from the one-good to the ten-good treatment and vice versa the next
hour. No participant took part in both auctions. Participation rates were 82%
(99 agreed to participate of 121 approached) for the one-good treatment, and
84% (93 of 111) for the ten-good treatment.
The Dealer treatment was similar to the one-good treatment except that a
monitor visited each dealer at his or her booth the night before the sports card
show. The monitor first gave each dealer an instruction sheet for the hypothetical
auction, and then administered the follow-up actual auction, on the promise they
would not leak any information to potential subjects. The treatment took about
two hours (6:30 pm to 8:30 pm), and the participation rate was 91% (30 of 33).
Let (biH k , biAk ) represent bidder i’s hypothetical and actual bids for the Cal
Ripken, Jr. Topps Traded card, where subscript H and A represent the hypothet-
ical or actual bid, k = one or ten represents the one-good or ten-good case, and
the superscript i is the bidder ranked in terms of their bid (i = 1, 2, . . . , n). Given
Auction design: case studies 233
these two institutions, the hypothetical and actual market prices could then be
compared to Beckett’s October 1995 book value for the cards (Cal Ripken, Jr.
Topps Traded card = $350).
Two measures of internal consistency of bidding were considered. First,
List and Shogren (1998) tested how other goods affect hypothetical and actual
bidding behavior. Bids for the Cal Ripken, Jr. Topps Traded card should be lower
in the ten-good case relative to the one-good case because values decline with
more substitution possibilities, holding the binding budget constraint constant
(i.e., the Le Chatelier principle). Second, dealer bids, those people with more
intense experience with market prices, will be more clustered than non-dealer
bids. Evidence from lab valuation suggests that as bidders gain experience with
the market and the going market price, the variability of bids declines as many
see the market price as an informative signal (e.g., Plott, 1996). Bids cluster
as bidders learn about the market within which they trade. One should reject
the hypothesis that the bid variability of dealers is similar to the variability of
non-dealers in the 1-good auction.
We consider three hypotheses to explore whether the context of choice
affects the calibration functions for the Cal Ripken Jr. card, bAone i
= f(bHone
i
),
bAten = f(bHten ), and bAdealer = f(bHdealer ). First, for all three calibration func-
i i i i
tions, we test the notion that no bias exists in hypothetical behavior, i.e., that
the regression intercept is zero and the regression slope coefficient on hypo-
i i
thetical bid bHone or bHten is one. Second, the no-bias hypothesis implies that
bids are symmetrical in that no extra bias exists when other goods are present
in the value elicitation process. We use a likelihood ratio test to determine if
the coefficients generated from the one-good treatment equal the coefficients
from the ten-good treatment. Finally, we test whether experience affects the
calibration function. Again a likelihood ratio test is used to determine if the
auction coefficients generated from non-dealers are the same as the coefficients
generated from dealers.
Consider the general pattern of bidding behavior in the three treatments.
Table 8.17 shows the mean and median bids and the demographic characteris-
tics for participants in the one-good, ten-good, and dealer treatments. A one way
ANOVA test indicates that the respective samples for the three treatments do not
differ by the socio-economic characteristics listed in Table 8.17, thereby assur-
ing that bids across treatments differ due to treatment rather than demographic
differences. The results show that the distribution of hypothetical bids lies to
the right of the distribution of actual bids. A Wilcoxon matched-pairs signed
ranks test rejects equality of the distributions at the 1% level for each treat-
ment (1-good: Z = −6.9; 10-good: Z = −7.7; Dealer: Z = −5.9). Additionally,
bias as revealed by mean central tendency shows the ratio of hypothetical-to-
actual overbidding ranges from 2.2 to 3.5, depending on auction type. This
level of overbidding is within the range of 1.0 to 10.0 observed in earlier work
234 Experimental Auctions
$91.71) and 68% ($125 to $40), and mean and median actual bids fell by 53%
($55.87 to $26.40) and 100% ($75 to $0); excluding zero bids, the median fell
by 53% ($75 to $35). Using a Wilcoxon test, we reject the hypothesis that the
populations for bids, hypothetical and actual, elicited in the one-good auction
were similar to bids elicited from the ten-good auction at the 5% level or better
(hypothetical: Z = −2.61; actual: Z = −2.06). Second, using a Moses test, we
also reject the equal variance hypothesis at the 1% level for hypothetical and
actual bids (hypothetical: M = 121; actual: M = 96). The dealers’ knowledge
of the common market value reduced the dispersion of their bids.
Table 8.19 presents estimates of the Tobit and OLS calibration functions,
i
bAone = f(bHone
i i
), bAten = f(bHten
i i
), and bAdealer = f(bHdealer
i
) across the three differ-
ent auctions. Models 1–2, 3–4, and 5–6 reflect the 1-good, 10-good, and Dealer
treatments. The OLS estimates suggest the calibration functions are concave
for both one-card auctions (dealer and non-dealer), suggesting the relationship
236 Experimental Auctions
Model
Swallow’s discussion of their calibration method, 1994); and the two bias func-
tion method used by Harrison et al. (1997): one to account for the downward bias
due to free riding and the other to account for hypothetical bias. The free riding
bias function is measured in a comparison of two real valuation situations for a
nature calendar, one of which features public provision of the calendar. While
intuitively appealing, the approach rests on little tested assumptions about the
transferability of bias functions between different contexts: real versus hypo-
thetical values, private versus public goods; and the untested presumption that
these biases are additive.
Good- and context-specific calibration suggests that one might be tempted to
skip over the hypothetical question and go directly to the actual auction. But for
policy debates over public goods, actual experimental auctions are constrained
by one’s ability to deliver the good. Randall (1997, p. 200) is more skeptical
about the general usefulness of calibration: “[t]he calibration issue, it seems to
me, is an audacious attempt to promote a Kuhnian paradigm shift. . . . I would
argue vigorously that the essential premise is unproven and the question is
therefore premature and presumptuous. The proposed new calibration paradigm
is at this moment merely a rambunctious challenger to the dominant external
validation paradigm.” On a good day, one could interpret his statement as a call
for more work on calibration. If so, future research in experimental auctions and
calibration should explore whether private goods can serve as reasonable proxies
for public good preferences, and the burden of calibration can be reduced by
clustering goods into a limited set of functions defined by the context of choice.
Calibration work has been a series of exercises in pattern recognition without
apology. One should not confuse pattern recognition with doodling. Rather
observations are a formal and organized model of the structure and functions of a
complex process: in this case, value formation. Researchers should be concerned
with Smith and Mansfield’s (1998, p. 210) admonition that lab valuation work
has thus far provided “little systematic treatment of how the circumstances
of choice influence the analyst’s ability to describe preference relations from
choice functions” – if, in fact, this claim were supported by the evidence.
But this is an overstatement. A review of the growing laboratory valuation
literature reveals some of the patterns observed about choice and valuation:
context matters (Kahneman et al., 1990; Buhr et al., 1993), the good matters
(Hayes et al., 1995, Fox et al., 1998), information matters (Melton et al., 1996;
Cummings and Taylor, 1999; Fox et al., 2002), exchange institutions matter
(Bohm, 1972, Rutström, 1998), market experience matters (Shogren, 1990,
Shogren and Crocker, 1994), price information sometimes matters (List and
Shogren, 1999), substitutes and complements matter (List et al., 1998). The
interested reader can find more examples in the literature.
The question remains why? Is it the exchange institution? The subject pool?
The good? The context? All the above? The answer is not immediately obvious.
Auction design: case studies 239
But it is unanswered questions like these that continue to make the gap between
intentions and actions a challenge in non-market valuation. This is probably
because the perception of a hypothetical stain has never really been systemat-
ically removed by an industrial-strength theory in over two decades of debate.
We agree with Mansfield’s (1998, p. 680) point that “the power of the calibra-
tion model could be improved by a better understanding of how individuals
answer [contingent valuation] questions, including the traits or attitudes that
inspire individuals to give more or less accurate answers.” No camp has a com-
pletely convincing and axiomatic explanation as to what creates or removes the
wedge between intentions and actions. And the lack of an analytical framework
increases the likelihood that this discussion will remain stagnate as a “did not,
did too” spat. The debate will likely be palliated only when a robust theoretical
or behavioral reason emerges as to why the wedge happens and whether it can
be controlled systematically.
Calibration research may eventually lead to generalizations about behavior
that convert experimental results into theory. Such a conversion requires that
we understand how and why the context of choice matters, as the literature on
experimental economics has revealed time and again. Restructuring calibration
models to include context-dependent preferences might be a place to start (see
Tversky and Simonson, 1993). If context matters, people might think about
values in both hypothetical and real terms, and the interaction of these two
representations might affect the gap between intentions and actions (see, for
example, Shafir et al.’s (1997) work on money illusion). Designing an experi-
ment to understand whether a model of context-dependent preferences can help
organize behavior in valuation exercises is a worthwhile next step.
an induced value setting. While Carson et al. (2000) focus on a voting refer-
endum mechanism for a public good, the idea underpinning the consequential
mechanism should be robust to any incentive compatible device.3 Shogren and
Tadevosyan (2006) implemented a second price auction in which the probability
the auction is binding is the treatment – real, hypothetical, or consequential.
Table 8.20 summarizes the design used by Shogren and Tadevosyan (2006).
An induced value, second price auction was used to test for rational bidding
behavior under consequential conditions. Let α represent the probability the
auction is binding, in which case the highest bidder buys the good at the second
highest bid; (1 − α) is the probability the auction is not binding, i.e., hypothetical
or inconsequential. Based on this probability the auction is binding, we create
three treatments: real (α = 100%), hypothetical (α = 0%), and consequential,
a positive probability the auction is binding (0% < α < 100%). For consistency
we follow the auction design of Shogren et al. (2001a), in which the sets of
induced values were randomly drawn from a uniform distribution on [$0.10,
$10.0] in 10 cent increments, the utilized induced values were (0.4, 1.8, 2.5,
3.2, 3.8, 4.0, 5.3, 6.1, 6.3, 6.5, 7.0, 7.6, and 8.4). Each participant was assigned
each value twice during the twenty rounds.
Standard second price auction experimental instructions were used, with the
additional explanation of the consequential auction. Before starting the auction,
the monitor read the experimental instructions and addressed any questions.
3 In a consequential voting model, no one person can believe he or she alone can actually change
the probability of the outcome – he or she just has to believe that there is some positive probability
the action will be implemented given his or her vote. This is the essence of the voter’s paradox –
no one vote matters on the margin, but all matter in the final tally. The probability of public good
provision is positive but exogenous to the voter. In theory, this exogenous probability makes the
vote consequential and provides the needed incentives to induce truthful responses in an incentive
compatible mechanism. Similarly, no bidder in the second price auction has to believe his or her
individual bid affects the probability the auction is binding to induce truthful bidding, only that
some exogenous positive probability exists that the auction will be binding.
Auction design: case studies 241
The bidders knew the winner was the highest bidder, and he or she paid the
second highest bid for the good. The experiment was run in three sessions with
twenty rounds; the probability, α, was as follows for the twenty rounds: 88%,
41%, 100%, 39%, 85%, 91%, 0%, 15%, 24%, 100%, 0%, 100%, 65%, 55%,
0%, 50%, 0%, 6%, 10%, 100%.
Each round had five steps:
Step 1: Each participant had a recording sheet showing his or her ID number,
induced value in that round (i.e., the price the monitor will pay to buy the good
back from the player if they win), the probability (0 ≤ α ≤ 100%) the auction was
binding in that round (identical for all bidders), and a place to write his or her bid.
Step 2: The bidders submitted their bids. Before doing so, we purposefully
told bidders about the potential risks from bidding too high or too low, i.e.,
pay too much or miss out on a profitable opportunity. The monitor revealed
the optimal bidding strategy information since the purpose is to test bidding
behavior in a consequential auction relative to the benchmark cases of real or
hypothetical auctions. Any deviation in bidding behavior from rationality or
treatment can be attributed to the notion of consequentialism, not a bidder’s
generic confusion about the best bidding strategy in the second price auction.
Step 3: The monitor collected and ranked bids from highest to lowest bid.
Step 4: The monitor announced whether the auction was binding based on a
random draw. The auction was binding if the random number was less than or
equal to the probability for that round. If the auction was binding, the highest
bidder was the winner, and he or she paid the second highest bid. The difference
between induced value and the second highest bid is the profit for the winner
for that round; no one else made any profits during that round.
Step 5: The next round started and Steps 1–4 were repeated. After twenty
rounds, the monitor paid the subjects their profits and a flat participation fee of
$10.
Table 8.21 summarizes bidding behavior for all rounds across treatments. The
unconditional results suggest the real auction induced more sincere bidding
than either the hypothetical or consequential auction. Mean bids were more
accurate, and the variation was smaller. In the real treatment, the mean bid minus
induced value was −0.08 (S.D. = 2.21); the mean bid minus induced value
was −$2.73 (S.D. = 5.25) for the hypothetical (outliers excluded) and −$0.91
(S.D. = 6.59) for the consequential.
Using conditional panel regression analysis, three results emerge. First, bid-
ders bid sincerely more frequently in the real second price auction relative to
either the hypothetical or consequential auctions. We test sincere bidding with
a two-way random effects model:
Hypothetical Hypothetical
Real (all obs.) (7 outliers Consequential
(α = 100%) (α = 0%) excluded)a (0% < α < 100%)
Parameter
Variable estimate Standard error Mean value Pr > |z|
N = 593
where Bidit is the bid submitted by ith player in tth round, β0 is the intercept
term, IVit is the induced value for the ith player in tth round, α1 equals 1 when
α = 100% (real), otherwise 0; α p equals 1 if 0 < α < 100% (consequential),
otherwise 0, γ i represents subject-specific characteristics, ϕ t represents round-
specific random effects, including learning or other trends in bidding behavior,
and εit is iid error. The interaction terms capture how slopes change for real
and consequential relative to the hypothetical behavior.
Table 8.22 shows the regression results. The joint hypothesis that the con-
stant was zero and the slope was one was tested to determine whether bidding
Auction design: case studies 243
I
Test 1 H1 : in hypothetical treatment 2 586 18.868 0.00008 Reject H0
the regression line has
intercept = 0, slope = 1
Test 2 H2 : in real treatment the 2 586 2.565 0.27739 Fail to reject H1
regression line has intercept
= 0 and slope = 1
Test 3 H3 : in consequential treatment 2 586 27.868 0.00000 Reject H2
regression line has an
intercept = 0 and slope = 1
II
Test 4 H4 : hypothetical and real 2 586 18.019 0.00012 Reject H4
behavior is the same
Test 5 H5 : hypothetical and 2 586 41.815 0.00000 Reject H5
consequential behavior is
the same
Test 6 H6 : real and consequential 2 586 7.845 0.01979 Could reject or fail
behavior is the same to reject the H6
behavior in each treatment differs from the optimal bidding, i.e., bid one’s
induced value:
H1 : β0 = 0; β1 = 1 (hypothetical)
H2 : β0 + β2 = 0; β1 + β4 = 1 (real)
H3 : β0 + β3 = 0; β1 + β5 = 1 (consequential)
Based on the Wald statistics, we cannot reject the H2 : bidding behavior in the
real treatment was not significantly different from the optimal bidding behavior
of truth telling (see Table 8.23, Part 1). The real auction is demand revealing. In
contrast, both hypotheses H1 and H3 are rejected; bidding in the hypothetical and
consequential treatments was significantly different than rational truth-telling
behavior.
The second main result is that bidding behavior was significantly different
across the auction treatments. Three additional hypotheses were tested:
H4 : β0 − β2 = 0; β1 − β4 = 0 (hypothetical vs. real)
H5 : β0 − β3 = 0; β1 − β5 = 0 (hypothetical vs. consequential)
H6 : β2 − β3 = 0; β4 − β5 = 0 (real vs. consequential)
Using a Wald statistic, all three hypotheses were rejected (see Table 8.23, Part 2).
Bidding behavior was not the same across treatments.
244 Experimental Auctions
Winner Price-setter
Real
n = 12 7 2 2 3
Percentage 58% 17% 17% 25%
Hypothetical
N = 12 1 1 3 0
Percentage 8% 8% 25% 0%
Consequential
N = 36 14 7 6 8
Percentage 39% 19% 17% 22%
The third primary finding is that the real second price auction had the great-
est percentage of highest-value winners and captured more of the potential
surplus than either the hypothetical or consequential auction. The bidder with
the highest value was more likely to win, and the bidder with the second high-
est value was more likely to set the price in the real treatments. Table 8.24
shows the highest bidder won 58% of the real auctions, but only 8% and 39%
of the hypothetical and consequential auctions, respectively. Measuring effi-
ciency by the percent of maximum surplus captured (max surplus = highest
induced value – second highest induced value) we see the real auction captured
153% of the potential surplus (see Table 8.21). The maximum surplus captured
in hypothetical treatment was negative (gains captured were overwhelmed by
the losses of winning bidders with low induced values); the surplus captured in
consequential treatment was positive but low, 29%.
While a consequential second price auction outperformed the hypotheti-
cal/inconsequential auction, it did not generate truth revealing bidding behav-
ior. Rather rational bidding behavior was more likely to be observed when the
auction was real (100% binding) relative to either the consequential or hypo-
thetical auctions. Most evidence in the extant literature finds people tend not to
state their real intentions in a hypothetical or inconsequential setting relative to
a real economic commitment.
The theoretical possibility we consider herein is whether bidding behav-
ior becomes more truth revealing if the auction is consequential, a positive
probability of being binding, as argued by Carson et al. (2000). These results
do not support that notion: bidding behavior in the consequential auction was
Auction design: case studies 245
neither optimal nor identical to real auction bidding. Rather bidding behavior
was truth revealing only if the auction was 100% binding. In response, one might
attempt to look for a consequential probability less than 100% that generates
the optimal bidding behavior. But based on past experience in the calibration
literature (see Fox et al., 1998), this search could end up being either good- or
context-specific or both, which would limit the general usefulness of the theory.
This search remains an open empirical question.
however, the evidence does not support the idea that a deadweight loss arises
in gift giving. Can one calibrate real and hypothetical bidding behavior using
ex post regressions? Yes, there is some correlation between what people say
they will do and what they actually do, but there is no general rule of thumb to
guide generic calibration. The calibration factors seem to be both good-specific
and context-specific. Finally, is the possibility that the valuation mechanism
is consequential – a chance the information gathered could influence policy –
sufficient to induce sincere bidding in a second price auction? No, the evidence
suggests that only a real auction (100% binding) was sufficient to trigger sincere
bidding behavior; consequential auctions (0% < binding probability < 100%)
did not induce sincere bidding.
The general lesson to take away from this chapter is that one need be prepared
to chase down the answers to many new questions piqued by your own curiosity
or by colleagues who raise legitimate questions about why you did what you
did. And while you do not have to run their untested experimental design for
them, you do have to support your decision of why you choose the particular
design route you did. These decisions can require auxiliary experiments such
as those presented in this chapter. The good news is that these new experiments
force you to think outside your normal spheres and will open doors to new
theories, new patterns, and new designs: all leading to deeper and more general
observations about human behavior.
9 Validity of experimental auctions
9.1 Introduction
This chapter addresses the question of whether experimental auctions are valid
in the sense that they accurately measure the theoretical notions of values laid
out in Chapter 3. Determining whether a measurement instrument is valid is a
complex concern for social scientists – made more difficult in that the theoretical
constructs we seek to measure are latent and unobservable. Validity refers to the
extent to which a measurement instrument actually measures what it purports
to measure. We must be clear about the object of measurement in experimental
auctions. Many psychologists have attacked traditional economists’ view of
preferences arguing that valuations are malleable, contextual, and even con-
structed on the spot (e.g., see Bettman, Luce, and Payne, 1998; Gunnarsson
et al., 2003). This argument is about the appropriateness of a given theoretical
model or construct; not with the validity of a measurement device per se.
We follow the economists’ notion of value as the theoretical construct of inter-
est (e.g., Chapters 2 and 3) and we seek to determine the validity of experimen-
tal auctions in measuring these latent variables. Our goal is to explore whether
experimental auctions provide an accurate measure of value and whether elicited
values respond in ways predicted by economic theory. Our focus on economists’
views of values and preferences serves as benchmark behavior against which
one can assess the validity of experimental auctions. We saw in Chapter 2 that
bidding behavior in induced value auctions generally conforms to what is pre-
dicted by auction theory: many studies find that bids in incentive-compatible
auctions are not statistically different than induced values. Unfortunately, the
real world might not be as stark or have values as well defined as in induced
value experiments. This chapter investigates the validity of experimental auc-
tions conducted with homegrown values; those values that people bring into
an experiment. This presents a unique challenge because we do not know true
homegrown values (if we did there would be no need for the study) and cannot
confirm whether bids match values. We must use indirect means of determining
the validity of experimental auctions with homegrown values.
247
248 Experimental Auctions
The psychology literature has identified several types of validity and has
defined a variety of validity tests (see e.g., Nunnally and Bernstein, 1994).
Although, most of the discussion in such literature is focused on determining
the validity of multiple-items scales that are formed from aggregating indi-
vidual responses to series of Likert “agree/disagree” statements, the general
principles are relevant here. Consider a variable with a true value given by xT
and a measurement of that variable, given by xM . The issue of validity can be
conceptualized by recognizing that xM = xT + es + er where es is systematic error
and er is random error. A measurement is valid if es = er = 0, which implies xM =
xT . A measurement is said to be reliable when er = 0. Reliability is a necessary
but not sufficient condition for validity. In large samples, if er is a mean-zero
random error (i.e., the measurement is reliable), a measurement is valid if there
is no systematic error.
We organize the findings regarding the validity of experimental auctions
around several key ideas. First, we investigate whether auction bids behave as
predicted by economic theory. Second, we examine whether auctions are reli-
able measures of value. Third, we consider the convergent validity of auctions.
Finally, we look at a three key behavioral anomalies.
the mean bid for two units of a good was less than twice than the mean bid
for one unit (e.g., people exhibit diminishing marginal utility), the mean bid
for two goods was higher than the mean bid for one good. For example, they
show that the mean bids for one bag of plain corn chips, USA branded corn
chips, and salsa were $0.51, $0.58, and $0.65, and the mean bids for two bags
of plain-labeled corn chips, American-labeled corn chips, and salsa were $0.95,
$1.07, and $1.13; the results pass the “scope test” as more is preferred to less.
List (2002) also finds that bids are increasing in the quantity of goods when
people simultaneously bid on smaller and larger bundles of sports cards. The
mean bid for non-dealers (dealers) for a bundle of ten sports cards was $3.72
($3.09) whereas the mean bid for a bundle of thirteen sports cards was $4.52
($3.45); however the effect disappeared when people viewed the bundles in
isolation rather than simultaneously, a finding we discuss later in this chapter.
Fourth, a concept closely related to diminishing marginal utility and down-
ward sloping demand curves is that economists expect demand to be affected
by the price/availability of substitutes and complements. Corrigan and Rousu
(2006c) investigated this issue by auctioning two substitute goods (plain-labeled
corn chips and American-labeled corn chips) and a complementary good (salsa)
in a variety of treatments that differed according to the quantity of each of the
goods that was available for sale. They found that, as predicted by economic
theory, bids were affected by the presence and availability of substitutes and
complements. For example, the mean bid to obtain both the American- and
plain-labeled corn chip bags was less than the sum of the bids to buy each
good individually (i.e., the goods were substitutes). By contrast, the mean bid
to jointly purchase American-labeled chips and salsa was $1.31, whereas the
sum of the bids to purchase each item individually was only $1.23. As predicted
by economic theory, demand was increasing in the availability of complements
and decreasing in the availability of substitutes. Cherry et al. (2004) also inves-
tigated this issue in an induced value auction. They showed that auction bids
were significantly influenced by the price of a non-auctioned good that was a
perfect substitute for the auctioned good. They found that people reduced their
auction bids when their value exceeded the price of the outside option and that
a decrease in the price of the outside option resulted in greater bid reduction.
Fifth, it would be expected that bidding behavior responds to positive and
negative information. Many of the goods that are sold in auctions can be char-
acterized by risk in that there is uncertainty regarding the quality or safety of a
novel auctioned good. Auction bids can be viewed as the certainty equivalent
of a lottery. The bid depends on a person’s belief or subjective probability that
the auctioned good will generate a particular outcome. This conceptualization
predicts that people will react to the introduction of new information by updat-
ing their prior probabilities concerning the likelihood of observing good and
bad outcomes. A number of studies have investigated how auction bids respond
Validity of experimental auctions 251
9.3 Reliability
We now investigate whether auctions generate reliable measures of value. Reli-
ability refers to the extent to which repeated measures of value relate to one
another. Reliability is a necessary but not sufficient condition for validity. A
common measure of reliability in psychology is a test-retest measure, where
the same person is given the same test at two points in time. To our knowledge,
Shogren et al. (2000) is the only experimental auction study to carry out a strict
test-retest procedure. In their study, each subject participated in four experi-
mental sessions over a two-week time period. In each session, subjects bid in
second price auctions to exchange an endowed: i) candy bar for another type of
candy bar, ii) non-irradiated pork sandwich for an irradiated pork sandwich, and
iii) apple for a mango. They found support for test-retest reliability as they could
not reject the hypothesis of equality of mean bids across all four sessions for the
candy bar and mango auctions; the only good for which bidding was not stable
across sessions was irradiated pork, the more exotic good. Although they found
little to no difference in mean bids across sessions, they did find that regardless
of the good, winning subjects significantly reduced their bids the next session;
Validity of experimental auctions 253
Round 1 1.00
Round 2 0.84 1.00
Round 3 0.76 0.94 1.00
Round 4 0.72 0.92 0.96 1.00
Round 5 0.71 0.89 0.93 0.97 1.00
a Data taken from second and random nth price auctions from
the no endowment treatment in Lusk et al. (2004); all correlation
coefficients are statistically different than zero at the 0.01 level of
significance.
an effect they attribute to preference learning where people bid higher to learn
how a new good fits in their preference set.
Other findings in the literature indicate that auctions are reliable: a) in auc-
tions with repeated rounds, bids are strongly correlated across rounds – this
is a quasi application of the test-retest method and is also related to internal
consistency reliability, b) different incentive compatible auction mechanisms
tend to generate similar measures of value in between-subject experiments,
and c) auction bids are reasonably stable across context (field versus lab) in
between-subject experiments.
Although still debated, it has been common practice to conduct auctions
over repeated rounds in which people submit bids for the same product(s)
after obtaining market information from the previous round. At issue here is
whether people are reasonably consistent in their bidding behavior across bid-
ding rounds. If people pulled random numbers from their head each round
to submit as bids, one would expect no correlation between bidding behavior
across rounds, but people’s bids tend to be strongly correlated across bidding
rounds. For example, Table 9.1 uses the data from Lusk et al. (2004a) and shows
the correlation coefficients between bids for Certified Angus Beef steaks across
five bidding rounds. All correlation coefficients are positive and statistically
different than zero; the lowest correlation is between the first and last bidding
rounds, but the magnitude of the coefficient (0.71) is still reasonably large indi-
cating a high degree of reliability. While Table 9.1 presents bidding behavior
for one good in one study, we observe a similarly high degree of consistency in
bids across rounds in virtually all repeated round auctions we have conducted.
Another type of reliability is called parallel-forms reliability which tests
whether different, parallel measures are highly related. In the psychology
254 Experimental Auctions
Random
Second price nth price English
auction auction auction BDM
a Data taken from the no endowment treatment in Lusk et al. (2004); all cor-
relation coefficients are statistically different than zero at the 0.01 level of
significance.
less uncertain about the value of the cookies in the store as other substitute
prices were readily available. People were more likely to make a purchase in
the store setting because they came to the store hungry.
than categorizing steaks into tough or tender categories, Fuez et al. (2004)
and Platter et al. (2005) used regression analysis to show that auction bids
were significantly decreasing in shear-force value (toughness). This finding is
noteworthy because people based their valuations only on the taste test – they
did not know one steak was objectively deemed more tender than another.
In a similar study, Umberger and Feuz (2004) also show that bids for steaks are
related to other palatability ratings obtained in the blind taste tests in anticipated
directions, e.g., people bid more on steaks they thought were juicer. Melton
et al. (1996a and 1996b) elicited bids for pork chops that varied by color, size,
and marbling level. Marbling refers to the amount of intramuscular fat content
in a cut of meat and higher marbling levels are associated with increased levels
of juiciness and tenderness. Melton et al. (1996b) found that after participating
in blind taste tests, people’s bids and market prices were increasing in the level
of marbling as would be expected. Platter et al. (2005) also had people bid on
steak of differing marbling levels. They found, as did Melton et al. (1996a), that
bids were increasing in the level of marbling (see also the studies by Killinger
et al., 2004a, 2004b, 2004c).
The ratio of average rating to the average bid was virtually equivalent across
varieties with a product category. At the individual level, just over 60% of sub-
jects have a weakly consistent ordering of varieties between the two measures,
in which a weakly consistent ordering occurs when the rating for product A
was greater than or equal to the rating for product B, the auction bid for prod-
uct A was also greater than or equal to the auction bid for product B. About
three-quarters of participants were consistent in identifying their most and least
preferred varieties with their bids and ratings.
Wertenbroch and Skiera (2002) elicited bids to obtain Coca-Cola and Cake
using a BDM mechanism. They found that bids for Coke and cake were sig-
nificantly correlated with responses to the statements “how thirsty/hungry are
you right now” and “How much do you like Coca-Cola/cake” in which people
responded on a 5-point scale (1 = not much; 5 = a lot). They further found
that the correlations between the auction bids and the two ratings were signif-
icantly higher than the correlations between hypothetical statements of value
and the two ratings. They state (p. 233), “As predicted, these results suggest
that WTPs from BDM provide a more valid measure of subjects’ preferences
than do WTPs from price matching.” They also showed that bids for ball-point
pens were significantly related to the attractiveness of the pen as determined by
responses to a nine-point scale.
Menkhaus (1992) did not specifically collect hedonic ratings in their beef
steak auctions, but they did collect similar information that should be related to
bids. For example, people that indicated a recent increase in poultry consump-
tion had lower auction bids for beef steak, people that were more concerned
about cholesterol bid less for beef steaks, and people that regularly froze steaks
prior to consumption placed a higher value on vacuum-packaged steaks over
steaks on over-wrapped styrofoam packages than people that did not typically
freeze steaks.
Like Menkhaus (1992), numerous studies exist that show auction bids related
to other hedonic-type information. For example, Lusk et al. (2001b) showed
that bids to exchange a bag of genetically modified corn chips for a bag of non-
genetically modified corn chips were increasing in concern for biotechnology
(as measured on a 10-point scale). Lusk et al. (2006) found that people’s bids to
avoid consuming a genetically modified food were increasing in the perceived
risks and decreasing in the perceived benefits of genetically modified food (as
measured on a 9-point scale). Overall, results from these studies imply strong
convergent validity between preferences expressed through auction bids and
preferences expressed through hedonic ratings and other “survey” measures of
preference.
At least one study, however, has not found strong correlation between hedonic
ratings taken after a taste test and subsequent auction bids. Jaeger and Harker
(2005) found that although responses to two different hedonic scales taken after
258 Experimental Auctions
tasting two Kiwi fruits were significantly correlated with each other, neither
were significantly related to bids for the fruit. Jaeger and Harker (2005) indicate
(p. 2523), “Clearly, the two types of data provide different information. Equally
clearly, the WTP data supplement the information gained through traditional
sensory and consumer testing.” These results indicate that while hedonic ratings
can be and frequently are related to auction bids, they need not be. Such a result
might arise when the hedonic ratings relate to preferences for one attribute, say
taste, whereas the auction bids encompass multi-attribute preferences, say taste
and visual appeal.
whereas WTP from the choice task significantly exceeded the stated price nor-
mally paid. Perhaps the most telling result is that WTP from the hypothetical
choice task was significantly related to the amount people normally paid for
pens whereas BDM bids were not. This led Wertenbroch and Skiera (2002)
to conclude (p. 237), “. . . this suggests that WTP is driven by subjects’ price
memories rather than their assessment of the value of the good in the current
purchase context when it is elicited under a hypothetical response format. It is
as if subjects start with what they normally pay and then adjust that estimate
upward to derive their WTP. In contrast, WTP under the BDM was predicted
by the weighted attractiveness . . . which suggests . . . that BDM leads subjects
to derive their WTP as a function of the perceived value of the good in the spe-
cific purchase situation.” More support for auctions comes from another part
of Wertenbroch and Skiera’s study which compared BDM bids for Coke and
cake to hypothetical open-ended statements of WTP. They found that whereas
people tended to round their hypothetical statements of value, auction bids were
more differentiated.
Wertenbroch and Skiera (2002) found that bids from an auction mechanism
(the BDM) were greater than WTP implied by a hypothetical stated choice task.
Their study does not permit one to isolate the effect of elicitation institution
(auction versus choice) and binding commitments. The inflation of valuations
in the choice task relative to the auctions could simply be due to the hypothetical
nature of the former. Lusk and Schroeder (2006) removed this confound and
compared non-hypothetical auction bids to non-hypothetical purchase decisions
using a between-subject design. In the auctions, people bid to obtain one of five
different types of beef steak and in the purchasing task, people chose one of the
five steaks, each of which was assigned a price (people were also allowed to
choose not to make a purchase at all). A series of discrete choices were made and
one was randomly selected as binding. Results indicated that valuations implied
from the choice task were significantly higher than auction bids (four different
auction mechanisms were used and the result persisted across all mechanisms).
This result is somewhat troubling as it implies that people frequently chose to
buy steaks at higher prices than what they would likely have bid in an auction.
One positive note is that the preference ordering of the five steaks was consistent
across the choice and auction elicitation mechanisms. Data from Lusk and
Schroeder (2006) show that preferences expressed through auctions and choices
were different, but they do not imply which is a “better” measure of value.
Frykblom and Shogren (2000) also compared auction bids for a book to non-
hypothetical purchase decisions to obtain the book at a given price. Unlike the
findings in Lusk and Schroeder (2006), they could not reject the hypothesis of
the equality of WTP from the auction and choice tasks.
Ding et al. (2005) also compared auction and choice data. Their investi-
gation focused on conjoint-type elicitation mechanisms. In a typical conjoint
260 Experimental Auctions
exercise people are shown a series of product profiles or descriptions that differ
along several attributes (e.g., brand, price, package size, etc.) and people are
typically asked to (hypothetically) rate or rank the desirability of each product
profile or to choose the one profile out of a set of profiles they most prefer.
They studied typical conjoint formats in addition to “incentive-aligned” (e.g.,
non-hypothetical) conjoint elicitation methods. In one treatment, people stated
their maximum buying price for each product profile, one of these profiles was
randomly chosen, and a BDM mechanism was used to determine whether the
profile was actually purchased. Another non-hypothetical treatment was similar
to Lusk and Schroeder (2006) in that people made a series of choices between
profiles and one of the choices was randomly selected as binding.
Ding et al. (2005) conducted a field experiment in a Chinese restaurant. Each
individual participated in one of four treatments in Part 1 of the experiment:
a hypothetical choice task, a non-hypothetical choice task, a non-hypothetical
stated price task that used the BDM bidding mechanism to determine whether
a purchase was made, or a hypothetical open-ended stated price task. After
completing this task, then regardless of treatment people moved to Part 2 of
the experiment and then made a single (non-hypothetical) choice of one Chi-
nese dinner out of twenty possible alternatives. Performance of the elicitation
mechanisms was judged primarily on the ability of preferences expressed in
Part 1 to predict the choice in Part 2. For clarity when Part 1 involved a non-
hypothetical task, a random draw determined whether the decisions in Parts 1 or
2 were actually binding. In the experiment, the treatment in Part 1 that involved
non-hypothetical choices was virtually identical to the decision task in Part 2;
the two tasks differed only in the options made available to people. The results
were that the two non-hypothetical tasks generated significantly better out-of-
sample predictions than the respective hypothetical tasks. The non-hypothetical
choice treatment showed the best out-of-sample predictive performance (it pre-
dicted the correct purchase 48% of the time as compared to the naı̈ve predictions
which would only be correct 5% of the time). This result is not terribly informa-
tive or surprising given that the hypothetical choice task exactly mirrored the
task used as the basis of performance. What is more interesting is the relative per-
formance of the other treatments: the non-hypothetical choice, non-hypothetical
stated price with BDM, and hypothetical stated price tasks correctly predicted
the Part 2 dinner purchase 26%, 15%, and 7% of the time. Ding et al. (2005)
attribute the finding that the non-hypothetical BDM mechanism did not per-
form as well as the hypothetical choice task to their design choice of screening
participants for interest in the study. They indicated (p. 74), “According to the
pilot study, it is almost certain that had we not screened out participants who
were not serious about the purchase decision the [non-hypothetical pricing task
with the BDM] would have performed better than the . . . hypothetical choice
conjoint.”
Validity of experimental auctions 261
9.5 Anomalies
We now discuss some findings that question the validity of experimental auc-
tions. To be more precise, the contention is not with experimental auctions
as a method to elicit values per se, but with the economists’ notion of prefer-
ences. Most economic models assume preferences are stable and independent of
context. But numerous studies stemming from psychology suggest malleable
preferences that are shaped or wholly constructed by frame and context. An
extreme view of wholly constructed preferences would argue against auctions
as a measure of value because such a view presupposes there is no “real” value
that exists to elicit. A more moderate view of contextual preferences would limit
the applicability of valuations obtained from auctions because such valuations
would only be “valid” in the context in which they were elicited and would
262 Experimental Auctions
with market experience as sports card dealers had similar value measures, but
inexperienced traders exhibited significant differences in WTP and WTA. Such
findings are generally consistent with the theoretical model put forth by Kolstad
and Guzman (1999), which implies that the WTP/WTA divergence can be
caused by costly information acquisition. Because people with greater market
experience have gained more information, one might expect WTP and WTA to
be more similar for such people according to the model in Kolstad and Guzman
(1999).
Zhao and Kling (2001) argued that if people are uncertain about the value of
a good and reversibility exists, WTP and WTA should diverge. Their “commit-
ment cost” theory suggests that the difference between WTP and WTA should
decrease when people are more certain of a good’s value, expect to obtain less
information about the good in the future, are more impatient, expect that revers-
ing the transaction becomes easier, and have more freedom in choosing when to
make the decision. Zhao and Kling (2001) reviewed much of the empirical lit-
erature on the WTP/WTA divergence and most of the findings were supportive
of the theory’s predictions. Kling et al. (2003) tested the theory by conducting
WTP and WTA auctions and measuring perceptions about the ease of delaying
the decision and reversing the transaction. When the perceived difficulty of
reversing the transaction was greater than the perceived difficulty in delaying
the decision, they observed the common result that WTA > WTP. When the
perceived difficulty in reversing the transaction was less than the perceived dif-
ficulty in delaying the decision, however, they actually found that WTP > WTA
– exactly what is predicted by the commitment cost theory.
Plott and Zeiler (2005) argue that the empirical evidence of the WTP/WTA
divergence is primarily due to subject misconceptions about the experiment.
Proper experimental control, which includes a variant on the BDM elicita-
tion procedure and extensive training with the elicitation mechanism, elim-
inate the WTP/WTA divergence. In particular, they show that they can turn
the WTP/WTA divergence “on” and “off” by varying experimental proce-
dures, a finding that suggests the endowment effect cannot be generated by
a fundamental feature of preferences. They conclude “. . . observed differ-
ences in WTP and WTA cannot be explained as an endowment effect based on
prospect theory and that broad implications based on such an interpretation and
advanced in the literature are inappropriate. Moreover, given the nature of our
results, claims that WTP-WTA gaps are unrelated to the experimental proce-
dures are clearly misleading with respect to the interpretation of the gap and its
implications.”
models based on standard expected utility theory with error best explain the
data.
Another interesting observation that has arisen is that people who reverse
preferences are subject to arbitrage in market settings. For example, suppose a
person chose lottery A over B but stated a monetary value of $3 for A and $5 for
B. The person could be sold B for their stated value ($5). Because the person
preferred A over B, however, they should be willing to give up B to have A.
Once in possession of A, the lottery can be bought back at the stated value of
$3. The end result is that the person has no lottery in their possession and has
lost $2. Chu and Chu (1990) and Cherry et al. (2003) have shown that when
people are vulnerable to arbitrage in a market-like setting they stop reversing
preferences. Chu and Chu (1990) show that people do not reverse preferences
even when they are not subject to arbitrage so long as they have previously been
exposed to arbitrage opportunities.
Cherry et al. (2003) showed that the rationality generated by arbitrage in
preference reversal experiments can “spill over” to non-market valuation set-
tings. Cherry and Shogren (2006) showed that rationality from such arbitrage
can “cross over” to other decision-making tasks. Another interesting observa-
tion from Cherry et al. (2003) is that after arbitrage, people do not change their
preference ordering but revise their stated values (bids) for high-risk lotteries
to conform to their choices.
Based on these findings Cherry et al. (2003) argue that rationality should not
necessarily be considered an individual-level phenomenon, but that markets
and other institutions induce rationality. If people have constraints on cognitive
abilities or must ration scarce cognitive resources, markets are necessary to
induce rationality. Markets can “create” rationality either by natural selection,
in which irrational people are forced out of the market due to bankruptcy or
by creating greater incentives for people to devote scarce resources to more
precisely articulate their preferences.
The good news is that experimental auctions naturally put people in a market-
like context and expose people to the discipline that markets impose. To the
extent that one believes preferences and values exist, auction markets provide
incentives for people to carefully consider and state their preferences.
In their first experiment, they asked people to first state whether they would
purchase six products such as wines, computer accessories, and luxury choco-
lates at a price equal to the last two digits of their social security numbers.
People were subsequently asked to state their maximum WTP for each item. A
random device was used to determine whether the first or second exercise was
binding; if the second was binding a BDM mechanism was used to determine
if purchase was actually made. At odds with economists’ view of preferences,
Ariely et al. found that the WTP bids were significantly correlated with social
security numbers; higher social security numbers led to higher bids. For exam-
ple, subjects with above median social security numbers bid from 57 to 107%
above people with below median social security numbers. Despite the apparent
malleability of absolute valuations, however, the relative valuations exhibited
remarkable stability as over 95% of subjects bid more on high rated wine than
low rated wine.
In a second set of experiments, Ariely et al. investigated valuations to avoid
listening to an annoying high-pitched sound delivered through headphones.
As in the previous experiment, people first answered a yes/no question about
whether they would listen to the sound for a given price (people were randomly
offered $0, $0.10, or $0.50) and then participated in a BDM mechanism in which
they bid an amount to avoid listening to the sound. The results suggest that bids
were influenced by the initial price indicating arbitrariness in preference; but the
relative valuations responded in intuitive ways, e.g., people increased valuations
when the sound was to last for longer periods of time.
List (2002) generated similar findings in a sports card market. He found that
the mean bid for a package of ten sports cards was significantly less than the
mean bid for a package of thirteen sports cards, when people viewed one and
only one of the packages. When viewed side-by side, however, people appeared
rational by stating higher bids for the package of thirteen sports cards than the
package of ten sports cards. This demonstration of a preference reversal shows,
as do the findings in Ariely et al., that relative valuations can appear rational
and coherent even if absolute valuations are not.
Ariely et al. argue that people may appear to behave rationally by avoiding
intransitive preferences so long as they remember all past transactions. Their
point is that the initial choice was not a “true preference” but was arbitrary
and influences all subsequent valuations. This argument in and of itself is not
terribly troubling for economists. Economics is silent about where preferences
come from. It only supposes that preferences exist and are not endogenous to
the problem at hand. Economics also has little to say about how preferences are
formed for a new good that people have never experienced, such as an annoying
noise. Only time will tell whether the apparent arbitrariness in preferences holds
up in market-like contexts in which people are subject to incentives and arbitrage
and whether the finding is isolated to “unfamiliar goods.” While List’s (2002)
Validity of experimental auctions 267
preference reversal result held up for card dealers and non-dealers, one would
be hard pressed to say that dealers are experienced in trading cards in bundles
of ten and thirteen.
It is also important to mention that several empirical findings suggest pref-
erences are less arbitrary than indicated in Ariely et al. In particular, stud-
ies that show “rational” changes in bidding behavior across treatments in
between-subject experiments would provide an indication that preferences are
not entirely constructed (e.g., Gunnersson et al., 2004). To that point, virtu-
ally countless contingent valuation studies have been published showing that
the frequency of people that would affirmatively vote for a policy is falling in
price – this is despite each person in such studies only receiving a single price.
These are between-subject demonstrations that demand curves are downward
sloping. It is also helpful to consider between-subject studies such as Rousu et
al. (2004a) which show that people who receive positive information about a
good bid more for the good than people who receive negative information. If
preferences were entirely arbitrary, valuations would be randomly distributed
in positive and negative treatments.
9.6 Summary
This chapter reviewed the literature to examine the question of the validity of
experimental auctions. Several studies have tested whether bidding behavior
conforms to basic tenets of economic theory, e.g. diminishing marginal utility
and downward sloping demand curves. Overall, the results support the notion
that bidding behavior is consonant with the comparative static predictions of
basic economic theory. Comparing bidding behavior with predictions of eco-
nomic theory is a test of convergent validity (a test that establishes whether
bids are related to other constructs they should be related to) and internal valid-
ity (tests of the cause-effect relationship between independent and dependent
variables).
Several studies have investigated convergent validity of auction bids in other
ways by comparing bids to preference ratings in taste tests, hedonic product
ratings, and other measures of value. The results suggest people tend to bid
higher for products they indicate as tasting better and for products given higher
hedonic rating scores. While the mean bid from experimental auctions differs
from mean willingness to pay inferred from non-incentive compatible contin-
gent valuation purchase questions, the value measures from the two approaches
are highly correlated. There is some conflict in the literature regarding whether
willingness to pay inferred from non-hypothetical choices is consistent with
auction bids; however, in the one study that found bids to be significantly lower
that willingness to pay from choice, the rank-ordering of goods was consistent
in both approaches.
268 Experimental Auctions
Only a few studies have investigated the external validity of auction values
by comparing whether auction bids correctly predict out-of-sample purchasing
behavior. In one such study, Ding et al. (2005) found that the BDM mechanism
performed three times better than a naı̈ve prediction and twice as well as a hypo-
thetical open-ended pricing task at predicting a subsequent non-hypothetical
choice. They also concluded that the BDM mechanism would likely have had
better predictive performance than non-hypothetical choices were it not for the
vagaries of their particular experimental design. In another study, Brookshire et
al. (1987) compared demand curves implied from bids in experimental auctions
to demand curves calculated from actual purchases of strawberries made via
door-to-door sales and were unable to reject external validity.
Taken together, these results lend strong support for the notion that experi-
mental auctions are valid measurement instruments and that the values elicited
in auctions are a valid theoretical construct. Some evidence exists to contradict
this view. For example, mean bids can differ across competing auction mecha-
nisms, all of which should be incentive compatible, a finding which undercuts
the reliability of the method. Such findings, however, only imply a weakness
in a particular implementation of the method or a weakness with one particular
mechanism, not necessarily with the entire valuation paradigm. There are other
“anomalies” presented as a challenge to the validity of experimental auctions
and economists’ notion of values. But repeated experiments and development
of economic theory have served to address some such challenges. In other
cases, we do not have enough information to understand under what condi-
tions economic theory and auction valuations “fail” and how economic theory
might evolve to address the concerns. Even in the instances in which violations
of individual rationality have been observed, other studies provide evidence
to suggest that the power of auctions and the discipline of markets have the
potential to generate rational behavior.
10 The future of experimental auctions
10.1 Introduction
Today researchers use experimental auctions to examine incentive and contex-
tual questions that arise in eliciting values through stated preferences methods.
The initial work developing the experimental valuation method was pioneered
decades ago by researchers interested in estimating the demand for public
goods, e.g., Bohm (1972), Bennett (1983), Knetsch and Sinden (1984), and
Cummings et al. (1986). Most experimental auctions follow a process simi-
lar to that described by physicist James Conant (1951; p. 56): “[a]bout three
centuries ago the trial-and-error experimentation of the artisan was wedded to
the deductive method of reasoning of the mathematician; the progeny of this
union have returned after many generations to assist the ‘sooty empiric’ in his
labors.” By combining pattern recognition with theoretical insight, the “sooty
empiric” uses experimental auctions to help clarify how incentives and context
affect how people state their preferences for real and hypothetical goods and
services.
This book has assimilated the current state of knowledge on the use of exper-
imental auctions to elicit values for goods. Our goal is to provide a resource
to practitioners interesting in designing and using experimental auctions in
applied economic, psychology, and marketing research. We have covered the
basics on value and auction theory to provide the analytical background for the
experiments. We then addressed specific issues related to design and imple-
mentation of experimental auctions. In many cases, we offered specific advice.
Other times, we provided guidance but no specific advice since there is too little
information in the extant literature to draw firm conclusions about the appropri-
ateness of a particular design issue. Implementation questions still exist, e.g.,
whether to use repeated bidding rounds with price feedback and whether to
endow subjects with a good prior to bidding remain debatable topics. Likewise,
while the general conclusion is that experimental auctions exhibit a high degree
of validity, several unresolved issues remain.
Uncertainties about the “appropriate design” or questions on validity arise
because research on experimental auctions is still in its early stages. Although
269
270 Experimental Auctions
a few research programs have explored the nature of experimental auctions for
nearly two decades, this is a relatively short period of time relative to that devoted
to other value elicitation research methods such as conjoint analysis and contin-
gent valuation. Only recently have papers on experimental auctions appeared in
outlets outside general and applied economics journals. For instance, Hoffman
et al. published their original work in Marketing Science in 1993, but almost ten
years passed until another paper appeared in a major marketing journal on the
topic (e.g., Wertenbroch and Skiera’s paper on the BDM mechanism appeared
in the Journal of Marketing Research in 2002).
Anecdotal evidence suggests researchers are increasingly interested in the
method as experimental auctions are discussed in work published in marketing
outlets. A recent paper by Ding et al. (2005), for instance, provides evidence
that marketing researchers should seriously consider using incentive compatible
value elicitation mechanisms. Also, experimental auctions are now showing up
in applied life science journals by non-economists. Examples of the method have
appeared in the Journal of Animal Science, Food Safety, the Journal of Muscle
Foods, the British Food Journal, the Journal of Food Quality and Preference,
Nicotine and Tobacco Research, Radiation Physics and Chemistry, and the
Journal of Dairy Science (see Table 1.1). While the use of experimental auctions
is increasing, we can always learn more as researchers apply the method in more
diverse contexts and for varied purposes.
contact with others who are making similar decisions in an exchange institution
puts in context the economic maxim that choices have consequences and stated
values have meaning for valuation and demand estimation. Relying on rational
theory to guide valuation and policy makes more sense if people make, or act as
if they make, consistent and systematic choices toward certain and risky events.
Experimental auction work should address the economic circumstances under
which the presumption of rationality is supported and when it is not.
The impact of rational valuation might be better understood if a person
could choose his preferred exchange institution, and if the researcher mea-
sures whether a person’s stated value is based on a complete and coherent set of
beliefs. Completeness implies a person’s risk neutral beliefs that measure value
are equal to the no-arbitrage price, if it exists (de Finetti, 1974). Coherence
or the no-arbitrage condition implies that the person does not accept a strictly
negative gamble, i.e., a sure loss or Dutch book. Implementing choice of an
exchange institution or some mix of institutions could be straightforward to
test in the lab. Let a person first select the auction or exchange institution that
he or she wants to participate in, and then elicit his or her value for the good in
question. Allowing for the choice of multiple exchange institutions complicates
a person’s decision problem, but it is necessary in a world of incomplete beliefs
about environmental values.
Second, do experimental auctions mimic real world behavior? Perhaps the
greatest impediment to increased use of experimental auctions in disciplines
outside economics is the dearth of studies testing the external validity of the
method. The following comments received by an anonymous reviewer of a
recent journal article emphasizes the point well, “My background is in psy-
chology . . . I do not believe that economists should be using such auctions to
study choice behavior or to elicit preferences that will be used to inform policy.
Until such time as economists undertake rigorous, systematic comparisons of
auction-revealed preferences and choices . . . in real markets, then people like
me . . . will continue to ask for evidence of external validity.” This comment
ignores the theory behind experimental auctions outlined in Chapters 2 and 3,
the extant empirical evidence for validity in Chapter 9, and the studies by Ding
et al. (2005) and Brookshire et al. (1987) that have tested for external valid-
ity. The comment also ignores that bids in experimental auctions are revealed
preferences obtained in a real market with real products and real money.
Still, this concern has some merit. The pertinent question is whether the
preferences revealed in an experimental auction market are related to prefer-
ences that govern other decisions in life such as daily shopping decisions. More
research is needed to determine the external validity of experimental auctions.
Forecasted market share or estimated demand curves obtained from experi-
mental auction bids might be compared to market shares and estimated demand
curves from retail sales data. Such findings would serve to determine whether
272 Experimental Auctions
the findings of Brookshire et al. (1987) and Shogren et al. (1999) are robust.
Relaxing the assumption of single unit demand and using multiple-unit vari-
ants of the Vickrey auction could be useful to elicit demand schedules for each
person.
Third, how well do experimental auctions forecast retail behavior relative
to other research methods? We need to understand better how our preference
measures help predict and explain various behaviors of everyday life. Examples
of behaviors of interest include the distribution of consumption within a fam-
ily, savings decisions, investment choices, and decisions about human capital
investment. Preference measures should be collected on the same people for
which we have extensive household survey data on choices. Even if one could
reject external validity of experimental auction bids could another method do
any better? A common approach in marketing studies is to construct a “hold-
out” choice as the last question in a survey/experiment and compare the ability
of different methods and models to forecast this hold-out choice, see, e.g., Ding
et al. (2005). The problem with using this approach to test external validity is
that while the hold-out choice is external to the initial decision making exer-
cise, it differs from a choice in a retail market in which individuals’ decisions
are not being scrutinized. It is not surprising that Ding et al. (2005) found that
a non-hypothetical choice task better predicted a subsequent non-hypothetical
choice than bids from a BDM mechanism because the hold-out choice was
essentially another replication of the previous choice-based decision task. A
more informative experiment would compare the performance of models esti-
mated from choice-based conjoint-type tasks to experimental auctions in terms
of their ability to predict market share and demand in a retail setting.
It is important to recognize that the goal of value elicitation is not always to
predict retail shopping behavior. One reason economists turn to primary data
collection methods such as experimental auctions is to measure externalities
that cannot be measured by analyzing retail purchases. The only feasible way to
measure externalities or the value of public goods is to put people in a decision-
making environment in which the public good or externality can be directly
traded. For example, consider the value a person places on a program provided
by public radio. It can easily be determined whether the person actually makes
donations to public radio, a practice akin to retail shopping behavior. Donations
to public radio in the “real world”, however, do not always reflect their true
value for the program. One reason is people might not donate because they
perceive their individual gift as having too small an impact on the viability of
public radio. Probably a more likely reason that donations do not reflect value is
that people can free ride off contributions of others, enjoying free public radio
without having to pay the price. The only way to truly measure a person’s value
for a program is to charge a price for the program and investigate whether the
person is willing to pay the price to listen. This is when an experimental auction
The future of experimental auctions 273
is useful. In an experimental auction, people could bid for the right to listen
to a program that would be otherwise unavailable. People have a dominant
strategy to bid their true value for the program; they can no longer free ride off
others’ contributions. Experimental auctions provide a measure of value in a
real market. Experimental auction markets may differ in terms of structure and
style from a traditional retail market, but they are real markets nonetheless.
Fourth, how do experimental auctions compare to choice-based value elicita-
tion methods and can the methods be combined to improve estimates of people’s
values? Auctions are not the only “game in town” when it comes to constructing
real markets. Some non-hypothetical choice-based methods are also incentive
compatible. It is important to realize that competing value elicitation methods
each have advantages and disadvantages. Experimental auctions have several
weaknesses including presenting a bidding task unfamiliar to some people, par-
ticipating in the auction can be cognitively burdensome, and the decision task
in an experimental auction (stating a bid) differs from shoppers’ routine tasks
of choosing a product at a posted price. Non-hypothetical discrete choice tasks,
in which people choose a product they wish to purchase at a given price, avoid
these weaknesses. Such discrete choice tasks are straightforward for most peo-
ple to answer because they mimic the natural environment.
Discrete choice tasks, however, have their own weaknesses. Discrete choices
only provide an approximate indication of preferences: one can only determine
whether one option is more preferred than another, but cannot necessarily deter-
mine how much more preferred one option is to another. People can be requested
to respond to repeated choice questions to gain more precision, but such a pro-
cedure entails a tradeoff between statistical precision, saliency, and participant
boredom. Further, when discrete choices are elicited one must assume some
functional form for utility and a form for the stochastic nature of the random
utility model such that choice data yields meaningful insight into willingness
to pay or market share. In contrast, auctions provide a point estimate of each
individual’s willingness to pay; auctions yield a continuous distribution of val-
uations independent of functional form assumptions.
No research method is a universal panacea. Auctions and discrete choice tasks
have their own relative advantages and disadvantages. More research is needed
to focus on combining the advantages of both methods to better characterize
preferences or to better forecast individual and market behavior. A body of
literature has begun to develop arguing for the combined estimation of revealed
and stated preference models, e.g., see Louviere et al. (2000) for a review, and
several empirical studies show that such joint models exhibit better performance
than either individual model (Azevedo et al., 2003). Such arguments are also
based on statistical literature showing that combined forecasts tend to outper-
form the forecasts of any individual model or individual, e.g., Clemen (1989).
Future research should focus on developing methods for combining auction
274 Experimental Auctions
and discrete choice data into a single prediction model. For example, perhaps a
model can be constructed in which auction data can be used to reduce the com-
paratively large variance in choice-based models that arises because discrete
choices only provide bounds on preferences. Future research might also seek
to identify conditions under which auction data might outperform choice data
and vice versa.
Fifth, can experimental auctions be more broadly employed to test economic
theory? One of the unique advantages of experimental auctions is that detailed
information can be gained about preferences without requiring strong a priori
assumptions about utility functionals. Experimental auctions are an excellent
method to use if one wishes to test the tenets of consumer theory in economics.
Chapter 9 discussed the results of several studies that have used experimental
auctions to perform such tests, but much more could be done. More research
is needed that focuses on using auction bids to estimate traditional systems
of demand curves used by economists (e.g., food, labor, health) even though
experimental auctions have rarely been used in such applications. More work
would be useful because most demand systems are estimated and consumer
theory tested using aggregate time series data even though the theory is meant
to hold at the individual level. Only when certain aggregation properties can
demand relationships be expected to hold at the aggregate level. Conditions
such as homogeneity and symmetry, commonly rejected in aggregate models,
can be more effectively tested in demand curves estimated from auction data.
Properties such as the weak axiom of revealed preferences (WARP) might also
be tested with auction data. Interest also lies in testing for alternative prefer-
ence structures, such as separability tests which focus on whether preference
rankings between two goods are affected by the price of a third good. Because
auctions can be used to construct individual-level demand curves, it might be
possible to discover significant heterogeneity in the degree to which different
consumers consider different bundles of goods as separable. Further, analy-
sis of demand systems frequently focuses on determining whether structural
changes in demand have occurred. If a panel of people routinely participated
in auctions, such tests of structural change could be more effectively tested.
Aggregate time series estimates of demand are plagued with concerns over the
endogeneity of price and quantity; experimental auctions suffer from no such
concern.
Sixth, can debates over design issues in experimental auction methods be
settled with non-experimental data? One advantage of additional research on the
external validity of auctions would be that such work would serve to solve some
of the controversies regarding the appropriateness of particular experimental
design issues. Arguments can be made for and against conducting repeated
bidding rounds with price feedback. Similarly, arguments could be made for
and against endowing participants with a good and eliciting bids to upgrade
The future of experimental auctions 275
formation processes, a conjecture which can be easily tested. This is the classic
Hawthorne effect: people are more productive when they know they are being
measured for their productivity. However, results from thought listing exercises
could serve as an initial way to look into decision makers’ black boxes.
Eighth, how do personality traits affect bidding behavior in experimental
auctions? It would be interesting to determine whether people with certain per-
sonality traits exhibit differences in bidding behavior compared to people with
a different set of traits. Such work might investigate whether personality has a
stronger influence on valuations or on how people react to the auction market.
One personality trait of direct relevance for experimental auctions relates to the
competitiveness of a person. Mowen (2004) found consumers are heterogeneous
in their degree of competitiveness. He related measures of competitiveness to
people’s interests in situations in which people directly compete against one
another (e.g., sports). Mowen also found that competitiveness affected prefer-
ences for products in which competition is vicarious (e.g., watching a drama
movie) and preferences for products that may be purchased for conspicuous
motives (e.g., a new innovation). Competition regards one’s standing relative
to others. Competitiveness is related to one’s concern for social standing and
the desire for social approval. For example, Ashworth et al. (2005) show that
people refuse to use coupons in the market place so as to not “look cheap,”
but will use coupons in settings in which they believe their actions will not
be perceived as “looking cheap.” Such findings have important implications
for experimental auctions and suggest new avenues for study. For example,
are more competitive people more likely to attend and participate in experi-
mental auctions? Do more competitive people bid higher in auctions because
they value the act of winning? Are certain products viewed as being more
competitive or conspicuous than others? Are people worried they might “look
cheap” if they bid too low? Research designs that vary the degree of anonymity,
the degree of information revealed about the distribution of individual bids, the
competitiveness of the mechanism, e.g., BDM versus second price auction, the
wording of instructions, e.g., using the term buyer instead of winner, might
provide interesting observations on this issue. One could measure personality
traits and competitiveness in pre- or post-surveys and use regression analysis
to identify the effect on bidding behavior.
Ninth, how do emotions and auction institutions interact? There remains an
open question about whether values are accentuated or attenuated by emotions
like the “fun of participation.” Several studies exist suggesting that emotions
can affect willingness to pay and willingness to accept valuations (e.g., Peters
et al., 2003; Lerner et al., 2004). For example Lerner et al. (2004) showed that
emotions induced by having people watch movie clips carried over to valuation
exercises – if people experienced the emotion of disgust, WTP measures of
value can exceed WTA measures of value, whereas the opposite was the case
The future of experimental auctions 277
for other induced emotions. The challenge is that economists usually work
by adding in only one emotion into a model as an extra degree of freedom;
they rarely if ever add two or three or more. If multiple emotions were added
to make homo economicus more human, either separability must be assumed
across emotions or assumptions must be made about how one emotion affects the
marginal productivity of another emotion (e.g., does envy increase or decrease
the marginal product of regret?). This challenge goes beyond what economists
usually know about behavior or have tried to address in modeling. Adding in one
emotion at a time or assuming separability is convenient but in the end rather
unsatisfying from the perspective of moving from homo economicus toward a
more human bidder.
Finally, can experimental auctions predict the future? A body of literature
has developed showing that prediction markets, where individuals buy and sell
contracts with values contingent on election outcomes, significantly outperform
polls in predicting election outcomes (e.g., Berg et al., 2006; Forsythe et al.,
1992). For example, in such a market, people buy and sell contracts that pay
$1 if Candidate A wins an election and $0 if Candidate B wins. Participants
in the market buy and sell the contracts depending on the expected success of
each candidate. If the “going price” of the contract is $0.60, this indicates that
the market predicts a 60% chance that Candidate A will win the presidential
election. Contracts can also be constructed that pay out the vote share of the
respective candidates, where the market prices can be interpreted as the market
consensus of the candidates expected vote share. Other studies have shown that
prediction markets are successful in forecasting the outcome of football games
(Servan-Schreiber et al., 2004), technological development (Pennock et al.,
2001), firm sales (Plott and Chen, 2002) and sales receipts from opening week-
ends of Hollywood movies (Wolfers and Zitzewitz, 2004).
The typical prediction market is a continuous double auction with buyers
submitting bids to buy (perhaps multiple) contracts and sellers submitting asks
to sell (perhaps multiple) contracts with transactions occurring when a bid
exceeds an ask at a price equal to the last bid or ask. The exchange institution
could be simplified to match the types of auctions considered in this book. For
example, a single contract that pays out $1 if an event happens prior to a given
date could be auctioned off in a second price auction. With such a procedure,
each person bids, and in so doing, provides an estimate of their value for the
contract. If risk neutrality can be assumed or induced through experimental
procedures, a person’s value for the contract should equal their expectation of
the probability that the event will happen before the given date. As described by
Davis and Holt (1993), risk neutrality can be induced by conducting the auction
in two stages where (1) earnings from the auction are determined in “points”
rather than money and (2) each person plays a lottery for a fixed monetary prize
where the chance of winning the prize is increasing in the number of points
278 Experimental Auctions
Ackert, L., B. K. Church, and G. P. Dwyer (2005) ‘When the Shoe is on the Other
Foot: Experimental Evidence on Evaluation Disparities’ Federal Reserve Bank of
Atlanta, Working Paper.
Addelman, S. (1962) ‘Orthogonal Main-Effect Plans for Asymmetrical Factorial Exper-
iments.’ Technometrics 4: 21–46.
Aldenderfer, M. S. and R. K. Blashfield (1984) Cluster Analysis. University Paper
Series on Quantitative Applications in the Social Sciences. Newbury Park, CA:
Sage Publications.
Alfnes, F. and K. Rickertsen (2003) ‘European Consumers’ Willingness to Pay for
U.S. Beef in Experimental Auction Markets’ American Journal of Agricultural
Economics 85: 396–405.
Ali, M. M. (1997) ‘Probability and Utility Estimates for Racetrack Bettors’ Journal of
Political Economy 85: 803–815.
Allenby, G. M. and P. E. Rossi (1999) ‘Marketing Models of Consumer Heterogeneity’
Journal of Econometrics 89: 57–78.
Ariely, D., G. Loewenstein, and D. Prelec (2003) ‘Coherent Arbitrariness: Stable
Demand Curves without Stable Preferences’ Quarterly Journal of Economics 118:
73–105.
Arnold, C. R. (1931) ‘The Place of Farm Accounting in Extension’ Journal of Farm
Economics 13: 57–64.
Ashworth, L., P. R. Darke, and M. Schaller (2005) ‘No One Wants to Look Cheap: Trade-
Offs Between Social Disincentives and the Economic and Psychological Incentives
to Redeem Coupons’ Journal of Consumer Psychology 15: 295–306.
Ausubel, L. M. and P. C. Cramton (1996) ‘Demand Reduction and Inefficiency in Multi-
Unit Auctions’ Working Paper, Department of Economics, University of Maryland.
Azevedo, C. D., J. A. Herriges, and C. L. Kling (2003) ‘Combining Revealed and Stated
Preferences: Consistency Tests and Their Interpretations’ American Journal of
Agricultural Economics 85: 525–537.
Baik, K. H., T. Cherry, S. Kroll, and J. Shogren (1999) ‘Endogenous Timing in a Gaming
Tournament’ Theory and Decision 47: 1–21.
Balistreri, E., G. McClelland, G. Poe, and W. Schulze (2001) ‘Can Hypothetical Ques-
tions Reveal True Values? A Laboratory Comparison of Dichotomous Choice and
Open-Ended Contingent Values with Auction Values’ Environmental and Resource
Economics 18: 275–292.
Bateman, I., A. Munro, B. Rhodes, C. Starmer, and R. Sugden (1997) ‘A Test of the
Theory of Reference-Dependent Preferences’ Quarterly Journal of Economics 112:
479–505.
279
280 References
Groves, R. M. (1987) ‘Research on Survey Data Quality’ Public Opinion Quarterly 51:
S156–S172.
Gul, F. (1991) ‘A Theory of Disappointment Aversion’ Econometrica 59: 667–686.
Gunnersson, S., J. F. Shogren, and T. L. Cherry (2003) ‘Are Preferences for Skewness
Fixed or Fungible?’ Economics Letters 80: 113–121.
Hammitt, J. K. and J. D. Graham (1999) ‘Willingness to Pay for Health Protection:
Inadequate Sensitivity to Probability?’ Journal of Risk and Uncertainty 18: 33–62.
Hanemann, W. M. (1984) ‘Welfare Evaluations in Contingent Valuation Experiments
with Discrete Responses’ American Journal of Agricultural Economics 66: 332–
341.
(1991) ‘Willingness to Pay and Willingness to Accept: How Much Can They Differ?’
American Economic Review 81: 635–647.
(1994) ‘Valuing the Environment through Contingent Valuation’ Journal of Economic
Perspectives 8: 19–43.
Hanley, N., J. F. Shogren, and B. White (2006) Introduction to Environmental Economics.
Oxford, UK: Oxford University Press.
Harless, D. W. (1989) ‘More Laboratory Evidence on the Disparity between Willing-
ness to Pay and Compensation Demanded’ Journal of Economic Behavior and
Organization 11: 359–379.
Harless, D., and C. Camerer (1994) ‘The Predictive Utility of Generalized Expected
Utility Theories’ Econometrica 62: 1251–1289.
Harrison, G. W. (1989) ‘Theory and Misbehavior of First-Price Auctions’ American
Economic Review 79: 749–762.
Harrison, G. W., R. Beekman, L. Brown, L. Clements, T. McDaniel, S. Odom, and M.
Williams (1997) ‘Environmental Damage Assessment with Hypothetical Surveys:
The Calibration Approach’ in M. Bowman, R. Brannlund, and B. Kriström (eds.)
Topics in Environmental Economics Amsterdam: Kluwer Academic Publishers.
Harrison, G. W., E. Johnson, M. M. McInnes, and E. E. Rutström (2003) ‘Individual
Choice in the Laboratory: Paradox Reloaded’ Working Paper, Dept. Economics,
University of Central Florida, September.
Harrison, G. W., R. M. Harstad, and E. E. Rutström (2004) ‘Experimental Methods and
the Elicitation of Values’ Experimental Economics 7: 123–140.
Harrison, G. W. and J. A. List (2004) ‘Field Experiments’ Journal of Economics Liter-
ature 42: 1009–1055.
Harstad, R. M. (2000) ‘Dominant Strategy Adoption and Bidders’ Experience with
Pricing Rules’ Experimental Economics 3: 261–280.
Hatcher, L. (1994) A Step-by-Step Approach to Using SAS for Factor Analysis and
Structural Equations Modeling Cary, NC: SAS Institute.
Hayes, D. J., J. F. Shogren, S. U. Shin, J. B. Kliebenstein (1995) ‘Valuing Food Safety
in Experimental Auction Markets’ American Journal of Agricultural Economics
77: 40–53.
Heckman, J. J. (2001) ‘Micro Data, Heterogeneity, and the Evaluation of Public Policy:
Nobel Lecture’ Journal of Political Economy 109: 673–748.
Heiner, R. (1983) ‘The Origin of Predictable Behavior’ American Economic Review 76:
560–595.
Hey, J. D. and C. Orme (1994) ‘Investigating Generalizations of Expected Utility Theory
Using Experimental Data’ Econometrica 62: 1291–1326.
286 References
Hoban, T. J., and J. Burkhardt (1991) ‘Determinants of Public Acceptance in Meat and
Milk Production: North America’ in P. van der Wal, G. M. Weber, and F. J. van
der Wilt (eds.) Biotechnology for Control of Growth and Product Quality in Meat
Production: Implications and Acceptability. The Netherlands: Pudoc Wageningen.
Hobbs, J. E., D. Bailey, D. L. Dickinson, and M. Haghiri (2005) ‘Traceability in the
Canadian Red Meat Sector: Do Consumers Care?’ Canadian Journal of Agricul-
tural Economics 53: 47–65.
Hobbs, J. E., K. Sanderson, and M. Haghiri (2006) ‘Evaluating Willingness-to-Pay
for Bison Attributes: An Experimental Auction Approach’ Canadian Journal of
Agricultural Economics 54: 269–287.
Hoffman, E., D. Menkhaus, D. Chakravarti, R. Field, and G. Whipple (1993) ‘Using
Laboratory Experimental Auctions in Marketing Research: A Case Study of New
Packaging for Fresh Beef’ Marketing Science 12: 318–338.
Hofler, R. and J. A. List (2004) ‘Valuation on the Frontier, Calibrating Actual and
Hypothetical Statements of Value’ American Journal of Agricultural Economics
86(1): 213–221.
Holt, C. A. (1986) ‘Preference Reversals and the Independence Axiom’ American Eco-
nomic Review 76: 508–515.
Holt, C. A. and S. K. Laury (2002) ‘Risk Aversion and Incentive Effects’ American
Economic Review 92: 1644–1655.
Hong, C. S. and N. Nishimura (2003) ‘Revenue Non-Equivalence between the English
and the Second-Price Auctions: Experimental Evidence’ Journal of Economic
Behavior and Organization 51: 443–458.
Horowitz, J. K. (2005) ‘The Becker-DeGroot-Marschak Mechanism is not Always
Incentive Compatible, even for Non-Random Goods’ Working Paper, Department
of Agricultural and Resource Economics, University of Maryland, March.
Horowitz, J. K. and K. E. McConnell (2000) ‘Values Elicited from Open-Ended Real
Experiments’ Journal of Economic Behavior and Organization 41: 221–237.
Horowitz, J. K., J. A. List, and K. E. McConnell (2006) ‘A Test of Diminishing Marginal
Value’ Economica, forthcoming.
Huber, J. and K. Train (2001) ‘On the Similarity of Classical and Bayesian Estimates
of Individual Mean Partworths’ Marketing Letters 12: 259–269.
Huck, S. and G. Weizäcker (2002) ‘Do Players Correctly Estimate What Others Do? Evi-
dence of Conservation in Beliefs’ Journal of Economic Behavior and Organization
47: 71–85.
Hudson, M. D., K. O. Coble, and J. L. Lusk (2005) ‘Consistency of Risk Premium
Measures’ Agricultural Economics 33: 41–49.
Huffman, W. E., M. Rousu, J. F. Shogren, and A. Tegene (2003) ‘Consumer Willingness
to Pay for Genetically Modified Food Labels in a Market with Diverse Information:
Evidence from Experimental Auctions’ Journal of Agricultural and Resource Eco-
nomics 28: 481–502.
(2004) ‘Consumers’ Resistance to Genetically Modified Foods in the US: The Role
of Information in an Uncertain Environment’ Journal of Agricultural and Food
Industrial Organization 2: Article 8.
Ibehndahl G., S. Isaacs, and R. Trimble (2002) ‘Financial Information Base of Partici-
pants in FSA Borrower Training’ Journal of Extension 40.
References 287
Mowen, J. C. (2004) ‘Exploring the Trait of Competitiveness and Its Consumer Behavior
Consequences’ Journal of Consumer Psychology 14: 52–63.
Mussa, M. and S. Rosen (1978) ‘Monopoly and Product Quality’ Journal of Economic
Theory 18: 301–317.
Nalley, L., D. Hudson, and G. Parkhurst (2004) ‘The Impacts of Taste, Location of
Origin, and Health Information on Market Demand for Sweet Potatoes’ Mississippi
State University Department of Agricultural Economics Research Report 2004–
001, July.
Neill, H., R. Cummings, P. Ganderton, G. Harrison, and T. McGukin (1994) ‘Hypo-
thetical Surveys and Real Economic Commitments’ Land Economics 70: 145–
154.
Neilson, W. (1994) ‘Second Price Auctions without Expected Utility’ Journal of Eco-
nomic Theory 62: 136–151.
Nevo, A. (2001) ‘Measuring Market Power in the Ready-to-Eat Cereal Industry’ Econo-
metrica 69: 307–342.
Norwood, F. B. and J. L. Lusk (2006) ‘The Dual Nature of Choice: When Personality and
Choice Collide’ Working Paper, Department of Agricultural Economics, Oklahoma
State University.
Noussair, C., C. Plott and R. Riezman (1995) ‘An Experimental Investigation of the
Patterns of International Trade’ American Economic Review 85: 462–491.
Noussair, C., S. Robin and B. Ruffieux (2002) ‘Do Consumers Not Care About Biotech
Foods or Do They Just Not Read the Labels?’ Economics Letters 75: 47–53.
(2004a) ‘Revealing Consumers’ Willingness-to-Pay: A Comparison of the BDM
Mechanism and the Vickrey Auction’ Journal of Economic Psychology 25: 725–
741.
(2004b) ‘A Comparison of Hedonic Rating and Demand-Revealing Auctions’ Food
Quality and Preference 15: 393–402.
(2004c) ‘Do Consumers Really Refuse to Buy Genetically Modified Food?’ The
Economic Journal 114: 102–120.
Nunnally, J. C. and I. H. Bernstein (1994) Psychometric Theory 3rd edn. New York,
NY: McGraw-Hill.
Organic Consumers Association. (2002) ‘Irradiation Home Page.’ http://
organicconsumers.org/irradlink.html
Parkhurst, G., J. F. Shogren and D. L. Dickinson (2004) ‘Negative Values in Vickrey
Auctions’ American Journal of Agricultural Economics 86: 222–235.
Pennock, D. M., S. Lawrence, C. L. Giles, and F. A. Nielsen (2001) ‘The Real Power of
Artificial Markets’ Science 291: 987–988.
Pennings, J. M. E., B. Wansink, and M. T. G. Meulenberg (2002) ‘A Note on Modeling
Consumer Reactions to a Crisis: The Case of the Mad Cow Disease’ International
Journal of Research in Marketing 19: 91–100.
Perry, M. and P. J. Reny (2002) ‘An Efficient Auction’ Econometrica 70: 1199–
1212.
Peters, E., P. Slovic, and R. Gregory (2003) ‘The Role of Affect in the WTA/WTP
Disparity’ Journal of Behavioral Decision Making 16: 309–330.
Phaneuf, D. J., J. A. Herriges, and C. L. Kling (1998) ‘Valuing Water Quality Improve-
ments using Revealed Preference Methods when Corner Solutions are Present’
American Journal of Agricultural Economics 80: 1025–1031.
References 293
Rozan, A., A. Stenger, and M. Willinger (2004) ‘Willingness-to-pay for Food Safety: An
Experimental Investigation of Quality Certification on Bidding Behaviour’ Euro-
pean Review of Agricultural Economics 31: 409–425.
Rutström, E. E. (1998). ‘Home-Grown Values and Incentive Compatible Auction
Design’ International Journal of Game Theory 27: 427–441.
Seip, K. and J. Strand (1992) ‘Willingness to Pay for Environmental Goods in Norway:
A Contingent Valuation Study with Real Payment’ Environmental and Resource
Economics 2: 91–106.
Servan-Schreiber, E., J. Wolfers, D. M. Pennock, and B. Galebach (2004) ‘Prediction
Markets: Does Money Matter?’ Electronic Markets 14: 2443–2251.
Shafir, E., P. Diamond, and A. Tversky (1997) ‘Money Illusion’ Quarterly Journal of
Economics 112: 341–374.
Shaw, W. D., R. M. Nayga, Jr., and A. Silva (2006) ‘Health Benefits and Uncertainty:
An Experimental Analysis of the Effects of Risk Presentation on Auction Bids for
a Healthful Product’ Economics Bulletin 4: 1–8.
Shin, S. Y., J. Kliebenstein, D. J. Hayes, and J. F. Shogren (1992) ‘Consumer Willingness
to Pay for Safer Food Products’ Journal of Food Safety 13: 51–59.
Shipman, D. R. (2001) ‘Marketing of Grains in Today’s Evolving Markets’ Paper pre-
sented at the meeting ‘Strategies for Coexistence of GMO, Non-GMO, and Organic
Crop Production,’ Minneapolis, MN, November 28.
Shogren, J. F. (1990) ‘The Impact of Self-Protection and Self-Insurance on Individual
Response to Risk’ Journal of Risk and Uncertainty 3: 191–204.
(1997) ‘Self-Interest and Equity in a Bargaining Tournament with Non-Linear Pay-
offs’ Journal of Economic Behavior and Organization 32: 383–394.
(2005) ‘Experimental Methods and Valuation’ in Handbook of Environmental Eco-
nomics. K. G. Mäler and J. Vincent, (eds.) Amsterdam: North-Holland.
Shogren, J. F. and T. Crocker (1994) ‘Rational Risk Valuation with Sequential Reduction
Opportunities’ Economics Letters 44: 241–248.
Shogren, J. and L. Tadevosyan (2006) ‘Bidding Behavior in a Consequential Second
Price Auction’ Working Paper, University of Wyoming.
Shogren, J. F., J. A. Fox, D. J. Hayes, and J. B. Kliebenstein (1994) ‘Bid Sensitivity and
the Structure of the Vickrey Auction’ American Journal of Agricultural Economics
76: 1089–95.
Shogren, J. F., S. Y. Shin, D. J. Hayes, and J. B. Kliebenstein (1994) ‘Resolving Dif-
ferences in Willingness to Pay and Willingness to Accept’ American Economic
Review 84: 255–270.
Shogren, J. F., J. A. Fox, D. J. Hayes, and J. Roosen (1999) ‘Observed Choices for Food
Safety in Retail, Survey, and Auction Markets’ American Journal of Agricultural
Economics 81: 1192–1199.
Shogren, J. F., J. A. List, and D. J. Hayes (2000) ‘Preference Learning in Consecutive
Experimental Auctions’ American Journal of Agricultural Economics 82: 1016–
1021.
Shogren, J. F., M. Margolis, C. Koo, and J. A. List (2001a) ‘A Random nth-Price Auction’
Journal of Economic Behavior and Organization 46: 409–421.
Shogren, J. F., S. Cho, C. Koo, J. List, C. Park, P. Polo, and R. Wilhelmi (2001b)
‘Auction Mechanisms and the Measurement of WTP and WTA’ Resource and
Energy Economics 23: 97–109.
References 295
297
298 Index
Christmas gift giving, welfare effects (case Dasgupta and Maskin mechanism 89
study) 225–228, 229, 230 data analysis
cluster analysis 108–109 censored regressions with auction bids
coherent arbitrariness 265–267 95–100
collective auction 69, 70 cluster analysis 108–109
commitment costs 43–44 elementary statistical analysis 95, 112
conditional mean regression, comparison with factor analysis 106–108
quantile regression 100–102 market share simulation 109–112
confounding factors 47–49, 50, 52–54 panel data regression with auction bids
conjoint analysis 4–5 103–106
consequential bidding (case study) 239–245 quantile regression with auction bids
context and control issues 53–54, 57–60 100–102, 103
balance between 174–175 demand, effects of price/availability of
balance in experimental auctions 6–16 substitutes and complements 79–80, 250
context-dependent preferences 238–239 demand reduction 53, 76–80
contingent valuation 4 demand revealing, performance of incentive
control and context issues 53–54, 57–60 compatible mechanisms 27–32, 33
balance between 174–175 diminishing marginal utility 76–80, 249
balance in experimental auctions 6–16 discrete choice models 4–5
controversial goods case study (demand for double hurdle model 98–100
GM food) 154–161, 163, 191–195 Dutch auctions, efficiency 28
conclusions 162–163
EU stance on GM food 154–155 eBay 62–63
experiment 155–157 econometric techniques 4–5
instructions for GM food auction 191–195 economic theory, evidence for predictions
results 157–161, 162 demand is affected by price/availability of
US stance on GM food 154–155 substitutes and compliments 250
use of biotechnology in food production 154 diminishing marginal utility 249
controversial goods case study (food from factors affecting WTP and WTA
animals treated with growth 251–252
hormones) 169–174 market price increases as demand increases
bovine somatotropin (bST) 169–170 248–249
conclusions 173–174 more is preferred to less 249–250
experimental design 170–171 testing reliability at individual level 5
health controversy 169–170 the value of a dollar to a person is exactly
porcine somatotropin (pST) 169–170 $1.00 252
results and discussion 171–173 values will rise (fall) with positive
controversial goods case study (irradiation of (negative) information 250–251
food) 163–169 see also auction theory
conclusions 167–169 economic value of choices 1
experiment 163–165 economists, use of information on people’s
results 165–167, 168 values 1–2
welfare effects of anti-technology messages efficiency of design 49–51
167–169 endowment effect 262–263
convergent validity 235, 255–261 case study 199–204, 205, 208, 209
auction bids and hedonic ratings 256–258 endowment versus full bidding 65–68
auction bids and other measures of value English auctions 17, 19–20, 24–25, 69
258–260 demand revealing performance 27–32, 33
auction bids and purchase decisions ex-post regression analysis 53
258–260 exchange institutions 1
auction bids and taste tests 255–256 expected utility theory 5
external validity 235, 261 relaxation of the independence axiom 24–25
cumulative prospect theory 42–43 experimental auctions
CVM-X method 237 active market environment 3–4
Index 299
advantage over other value elicitation study setting and context (field versus
methods 3–5 laboratory) 57–60
applications 6, 7–14 use of students as subjects 46–47
control/context balance 6–16 experimental design 47–54
description of heterogeneity in valuations ABA and BAB design 52–53
4–5 aliased effects 49, 50
determination of individual willingness to assignment of units to treatments 48, 51–52
pay 4 balanced design 49–51
early work 5 common expectation among participants 53
elicitation of homegrown values 6–16 confounding factors 47–49, 50, 52–54
exchange mechanism 3–4 control of design variables 47–51
incentive compatible mechanisms 3–4, definition of experimental unit 51–52
16–17 demand reduction 53
induced value experiments 5–6, 15–16 efficiency of design 49–51
purpose 5 ex-post regression analysis 53
two basic strategies 16 extraneous variables 52–54
valuation of non-market goods 3–4 finding designs 51
valuations are directly obtained 3–4 fractional factorial design 48–51
see also English auctions; BDM full factorial design 47–48, 49
mechanism; collective auctions; Dutch issues of control and context 53–54
auctions; nth price auctions; random nth main-effects only design 48–51
price auctions; second price auctions; orthogonal design 49–51
Vickrey’s second price auctions randomization 48, 52
experimental auctions (conducting) replication of treatments 51–52
affiliation of values in repeated bidding software 51
rounds 80–88, 90, 92 within-subject design 48, 52–53
avoiding misperceptions in participants experimental unit, definition 51–52
62–65 external validity 261
BDM mechanism 69–70
best practices 62 face validity of data 6–16
choosing an auction mechanism 69–76 factor analysis 106–108
collective auction 69, 70 farm financial records valuation (case
Dasgupta and Maskin mechanism 89, 90 study) 149–154, 186–190
demand reduction 76–80 benefits of farm recordkeeping 149–150
diminishing marginal utility 76–80 conclusions 152–154
endowment versus full bidding 65–68 data and methods 150–151
English auction 69 instructions for financial records auction
field substitutes 79–80 186–190
focus groups 62 results 151–152, 153
initial qualitative study 62 field substitutes, effects of 79–80, 250
learning in repeated bidding rounds 80–88, ‘first choice’ or ‘highest utility’ rule
90, 92 110–111
multiple good valuation 76–80 first price auction 23–24
negative values 92–94 efficiency 28
nth price auction 69 fixed or fungible preferences (case
random nth price auction 69, 70 study) 217–222, 223, 224, 225
repeated bidding round auctions 80–88, 90, focus groups 62
92 food from animals treated with growth
second price auction 69 hormones (case study) 169–174
training and practice for participants bovine somatotropin (bST) 169–170
62–65 conclusions 173–174
experimental auctions (preliminaries) experimental design 170–171
experimental design 47–54 health controversy 169–170
sample size determination 55–57 porcine somatotropin (pST) 169–170
study objectives 46–47 results and discussion 171–173
300 Index
preference learning for unfamiliar goods (case sample size determination 55–57
study) 196–199 comparison of means from two independent
preference reversals 263–265 samples 55–56
case study 217–222, 223, 224, 225 distribution of a valuation in the population
preferences 56–57
construction 218 second price auctions 16–17, 19–20, 69
fixed or fungible (case study) 217–222, 223, demand revealing performance 27–32, 33
224, 225 non-expected utility preferences 24–25
stability (case study) 217–222, 223, 224, see also Vickrey’s second price auction
225 second price auction tournaments (case
price discrimination models, effects of study) 209–214, 215, 217
valuation heterogeneity 5 sellers, willingness to accept 1
psychologists, use of information on people’s share of preference model 111
values 1–2 stated preference methods 2–3
public policy, determination of welfare effects unreliability of values elicited 3
5 see also informing policy case studies students, use as subjects 46–47
purchase decisions, comparison with auction study objectives 46–47
bids 258–260 study setting and context (field versus
laboratory) 57–60
quantile regression with auction participants influenced by being watched 60
bids 100–102, 103 substitute availability, effects on
censoring 101, 102 demand 79–80, 250
comparison with conditional mean
regression 100–102 taste tests, comparison with auction
bids 255–256
random nth price auctions 17, 19–20, 69, 70 test-retest reliability 252–253
demand revealing performance 27–28, Tobit model 97–100
30–32, 33 tournament auction
see also nth price auctions comparison with standard auction (case
random parameter models 4–5 study) 209–214, 215, 217
randomization 52 demand revelation (case study) 209–214,
rank-dependent expected utility theory 24–25, 215, 217
42 effects on insincere bidding (case study)
reliability 209–214, 215, 217
consistency across context 254–255
consistency across repeated rounds 253 unengaged bidders
definition 252 and auction mechanism (case study) 202,
of experimental auction measurements 206–208, 209
252–255 effects of tournament auction (case study)
parallel-forms reliability 253–254, 255 209–214, 215, 217
relation to validity 252 unfamiliar goods, preference learning (case
test-retest 252–253 study) 196–199
repeated bidding round auctions 80–88, 90,
92 validity
affiliation of values 80–88, 90, 92 definition 247
consistency across rounds 253 external 235, 261
learning in 80–88, 90, 92 validity of experimental auctions 247–248
replication of treatments 51–52 anomalies 261–267
revealed preference methods, implicit auction bids and economic theory 248–252
values 2–3 coherent arbitrariness 265–267
risk, definition 37 convergent validity 235, 255–261
risk aversion 39–41 endowment effect 262–263
risk perception 39–41 measurement of homegrown values
risk preference 39–41 247–248
risk premium (WTP to avoid a risky good) 40 object of measurement 247–248
Index 303